OpenAI Expands Daybreak With GPT-5.6-Cyber, Adding Daybreak Blue and Daybreak Red Tiers

OpenAI has expanded its cybersecurity defense programme, Daybreak, introducing two access tiers- Daybreak Blue and Daybreak Red, aimed at giving approved defenders greater access to frontier AI capabilities for security work. Daybreak Blue offers access to general-purpose models, including GPT-5.6 Sol, tailored for tasks such as vulnerability discovery, malware analysis and incident response. Daybreak Red goes further, offering access to a new specialised model, GPT-5.6-Cyber, built specifically for advanced vulnerability research and exploit validation, with significantly fewer refusals on complex, high-risk security tasks compared to its predecessor. OpenAI said the new model has already been used to uncover previously unknown, high-severity vulnerabilities in real-world software, including Google's Chrome browser, a popular mobile operating system and a widely used database, with findings responsibly disclosed to the respective companies. Access to both Daybreak tiers remains restricted to vetted individuals and organisations, with OpenAI introducing additional safeguards including mandatory hardware security keys from September 2026. The move follows Anthropic's earlier launch of its own cybersecurity-focused model, Mythos, as AI labs increasingly position themselves at the centre of enterprise cyber defense.

OpenAI on Monday said it is expanding Daybreak, its exclusive cybersecurity initiative, to help give its participants access to the "right capabilities" that they need to help defend themselves from attackers. The company first introduced Daybreak in May, shortly after its chief rival Anthropic captivated Wall Street and the U.S. government by launching its own cybersecurity coalition called Project Glasswing. OpenAI positioned Daybreak as a way for its ecosystem partners to use its most advanced artificial intelligence models to adapt to a rapidly changing threat landscape. OpenAI said Monday that it is expanding the program to include two different access tiers: Daybreak Blue and Daybreak Red. The expansion follows a number of cybersecurity incidents that AI developers —including OpenAI, Anthropic and Meta — have disclosed in recent weeks. In each case, an AI model accessed systems that should have been off limits as part of cybersecurity testing, prompting industry researchers and government officials to call for stronger protections. "As the threat landscape evolves, we're putting frontier intelligence in the hands of trusted defenders before attackers can deploy offensive AI at scale," OpenAI said in a post on X on Monday. Daybreak Blue will give users unique access to OpenAI's advanced general-purpose models, as the safeguards will be altered to allow for defensive security work, the company said in a blog post. Daybreak Red participants will be able to go a step further, leveraging OpenAI's "purpose-trained cybersecurity models" for security testing, vulnerability research and exploit validation. OpenAI is also launching a new AI model, GPT‑5.6‑Cyber, which will be available to Daybreak Red users. The company said the model is built on its most powerful publicly available offering, GPT-5.6 Sol, but it's designed to improve capabilities and reduce refusals on certain specialized cybersecurity tasks. The company said it recommends Daybreak Blue as the starting point for most organizations. OpenAI has repeatedly warned about AI models' advancing cyber capabilities in recent months. The company said last week that it is pausing some internal activities involving an upcoming model called Astra, which has demonstrated "significant advancements in agentic coding and cybersecurity" during testing. It said it is working to assess these capabilities and implement more robust safeguards and security controls. "We're committed to working alongside governments, safety institutes, and civil society to ensure that the frontier capabilities of models like Astra, and those that follow, are deployed responsibly and broadly for the benefit of all humanity," OpenAI said.

OpenAI announced Monday it was expanding access to its frontier models for defensive cybersecurity, detailing different defensive and red-teaming workflows and a new partner program with major cybersecurity product providers. In a pair of blogs posted Monday, OpenAI said it was updating its Daybreak program – which provides unreleased frontier models to private organizations and governments for defensive cybersecurity work – and introducing a new model variant. Daybreak Blue, powered by OpenAI’s ChatGPT-5.6-Sol, would operate with lower cybersecurity safeguards compared to other commercially available models and is described as “a recommended starting point for most defenders” that supports tasks like vulnerability discovery, secure code review, malware analysis, incident response and patch validation. Daybreak Red, meant for more advanced red-teaming, would provide access to a new model, dubbed GPT-5.6-Cyber, that the company said is more purpose-trained for finding vulnerabilities and testing (or exploiting) them. The model is also less likely to refuse requests around “dual-use cyber tasks.” According to OpenAI, the organizations in Daybreak Red will have their use closely monitored and supervised, as GPT-5.6-Cyber is significantly more capable in carrying out malicious cyber tasks than Sol. A security evaluation the company devised tested both models on complex requests, including exploit chain development, authentication bypass, privilege escalation and other hacking tasks. Sol succeeded in 1.5% of the requests, while Cyber completed 95%. OpenAI said it plans to publish a more detailed system card for GPT-5.6-Cyber at a later date. “Models running with reduced safeguards carry risks beyond standard model usage, whether from misuse or misalignment,” the company said in a blog. “Despite these risks, we believe that democratizing access to frontier intelligence for defenders is crucial to accelerating and automating cyber defense.” Additionally, OpenAI announced a partnership program with 16 major cybersecurity providers, saying organizations could access their models through their existing security services. The partners include IBM, CrowdStrike, Accenture, Ernst & Young, KPMG, Palo Alto Networks, Cisco, Cloudflare, Sophos and others. “These partners bring deep security expertise and established relationships with organizations around the world,” OpenAI said in its blog. “By bringing our frontier cyber models into their services, we can help more defenders find serious vulnerabilities, validate which ones matter, and fix them faster.” Companies like OpenAI, Anthropic and others are trying to rebalance their priorities after a string of AI-agent sandbox escapes have rattled policymakers and caused some cybersecurity experts to question if AI companies are doing enough to properly isolate the models from the internet during testing. Last week, OpenAI said it was intentionally slowing down development of its newer “Astra” model in order to develop better guardrails to restrain its behavior. Cybersecurity and AI experts have told CyberScoop that while AI systems have greatly improved at finding and exploiting vulnerabilities in software code, they still require substantial human guidance and supporting infrastructure to operate as intended. Additionally, some research has shown that without such guidance, even near-frontier models can struggle to fully patch a discovered vulnerability or avoid introducing new bugs with their fixes.

Image of the OpenAI logo. - Europa Press/Contacto/Omar Marques MADRID, Aug 11. (Portaltic/EP) - OpenAI has announced the launch of GPT-5.6-Cyber, its new model specialized in advanced cybersecurity, and the expansion of Daybreak, its global cybersecurity program for detecting and fixing vulnerabilities for organizations, which will now be available with Daybreak Blue and Daybreak Red access levels. GPT-5.6-Cyber has been developed based on GPT-5.6 Sol, released last month, though with differences focused on its specialization for tasks in advanced security research, vulnerability development, and attack testing. In fact, as explained by the Artificial Intelligence (AI) lab on its website, this new model has been tested with benchmarks such as ExploitGym, which aims to assess whether agents can convert vulnerabilities into functional exploits. In the results, GPT-5.6-Cyber outperformed both GPT-5.6 Sol and GPT-5.5-Cyber. Continuing along this line, one of the areas in which OpenAI has strengthened GPT-5.6-Cyber is its ability to find and precisely calibrate the severity of new zero-day vulnerabilities, that is, a security flaw unknown to developers and with no available solution upon discovery. Furthermore, compared with GPT-5.6 Sol, GPT-5.6-Cyber completes 95 percent of advanced and dual-use cybersecurity requests, while Sol rejects almost all of them, completing only 1.5 percent. For example, OpenAI researchers have stated that they used GPT-5.6-Cyber to identify previously unknown vulnerabilities in V8, Chrome’s JavaScript engine, which could be chained to corrupt memory and escape the heap sandbox. The company has also noted that when it launched its frontier model GPT-5.6 Sol, it included a set of cybersecurity safeguards. These safeguards are now lifted with the Daybreak program, which expands availability to more organizations. TWO DAYBREAK ACCESS LEVELS: BLUE AND RED Specifically, the company has expanded Daybreak access by dividing it into two modalities with different verification requirements. On one hand, it announced Daybreak Blue, an access level tailored for routine defensive work that companies or organizations may require, such as code review or malware analysis. This level provides access to general models like GPT-5.6 Sol, while removing the restrictions that apply to models available to all users. On the other hand, it launched the Daybreak Red level, which offers advanced access to cutting-edge models for researchers and testing teams, including the new specialized model GPT-5.6-Cyber. However, OpenAI clarified that access to these levels is limited to "approved individuals and organizations" conducting authorized work, with identity verification, account security, monitoring, approved usage restrictions, and legal declarations. In this regard, OpenAI is expanding Daybreak through security and services partners, including Accenture, Cisco, Cloudflare, CrowdStrike, IBM, Palo Alto Networks’ Unit 42, and others. "This enables more organizations to have a pathway to use cutting-edge cybersecurity models through trusted products, managed services, and client collaborations," it added. Through the program, OpenAI has stressed that community initiatives are created, referring to partnerships with security firms as well as with the open-source community, and AI-assisted review is encouraged to identify and fix flaws.

El lanzamiento de GPT-5.6-Cyber completa una reorganización en el programa Daybreak de OpenAI. Esta iniciativa de ciberseguridad ahora se divide en dos niveles de acceso: Blue y Red. El objetivo declarado es dar a los defensores más margen frente a ataques impulsados por inteligencia artificial. En un momento en que estos modelos son cada vez más capaces de encontrar y explotar vulnerabilidades por sí solos, esta estrategia cobra relevancia. Daybreak nació en junio junto a GPT-5.5-Cyber, con acceso restringido para investigadores de seguridad seleccionados. Ahora esa estructura se reparte en dos vías diferenciadas según el tipo de trabajo que se necesite hacer. Lanzamiento de GPT-5.6-Cyber por OpenAI: diferencias entre Daybreak Blue y Red Daybreak Blue da acceso a los modelos generalistas de OpenAI, incluido GPT-5.6 Sol, con salvaguardas adaptadas para trabajo legítimo de seguridad. Según la compañía, es la opción pensada para la mayoría de desarrolladores. Cubre tareas como: - Descubrimiento de vulnerabilidades - Revisión segura de código - Análisis de malware - Respuesta a incidentes - Validación de parches Daybreak Red, en cambio, está reservado para investigación de vulnerabilidades, validación de exploits y pruebas de seguridad más agresivas. Es aquí donde vive GPT-5.6-Cyber. Este modelo se construyó sobre GPT-5.6 Sol pero con menos restricciones frente a peticiones de doble uso que el modelo normal rechazaría, como encontrar vulnerabilidades de día cero o desarrollar cadenas de exploits. El salto de rendimiento que justifica el cambio de nivel Para medir esta capacidad, OpenAI ha introducido una evaluación llamada Advanced Cybersecurity Completion Rate. En ella, GPT-5.6-Cyber completó el 95% de las peticiones avanzadas. Estas incluyen cadenas de exploits, bypass de autenticación y escalada de privilegios. GPT-5.5-Cyber se quedó en el 57,3%. Por tanto, la mejora es significativa entre versiones. GPT-5.6 Sol sin ajustes apenas llegó al 1,5%. Sin embargo, este porcentaje sube al 2% cuando se utiliza a través de su versión Blue en el programa Daybreak de OpenAI. La diferencia entre ambos niveles deja claro que el acceso Red no es una variante cosmética del mismo modelo. De hecho, es una herramienta pensada para tareas que el modelo generalista está diseñado para evitar. OpenAI asegura haber usado GPT-5.6-Cyber para encontrar vulnerabilidades reales antes desconocidas. Incluidas dos en el motor V8 de Chrome que permitían corromper memoria y escapar de su sandbox. Google ya las ha corregido. La compañía afirma también haber identificado cientos de fallos adicionales en sistemas operativos de escritorio, móviles y bases de datos. OpenAI ha insistido de nuevo en que GPT-5.6-Cyber no tuvo relación con el ataque sufrido por Hugging Face. Esta aclaración se repite tras el fin de semana en el que la propia compañía reconoció que su próximo modelo Astra podría alcanzar el nivel crítico de capacidad ciberofensiva dentro de su marco de preparación. Puedes seguir a HardwarePremium en Facebook, Twitter (X), Instagram, Threads, BlueSky o Youtube. También puedes consultar nuestro canal de Telegram para estar al día con las últimas noticias de tecnología. FAQ Solo investigadores de seguridad aprobados por OpenAI, ya que este nivel da acceso a modelos con menos restricciones frente a peticiones de doble uso como busqueda de vulnerabilidades de dia cero. Si, es su sucesor directo dentro de Daybreak Red y mejora notablemente su tasa de exito en tareas ofensivas avanzadas, del 57,3% al 95% segun la evaluacion interna de OpenAI. No. OpenAI ha negado explicitamente esta relacion en el mismo anuncio del modelo, tras las especulaciones surgidas el fin de semana previo. Es el proximo modelo de la compañia, que segun reconocio la propia OpenAI podria alcanzar el nivel critico de capacidad ciberofensiva dentro de su marco interno de preparacion.

OpenAI unveiled its new model specialized in advanced cybersecurity, named GPT-5.6-Cyber, together with the expansion of its global Daybreak program, which will now offer two access levels: Daybreak Blue and Daybreak Red. This initiative aims to facilitate the detection and remediation of vulnerabilities in organizations worldwide. GPT-5.6-Cyber was built on the foundation of GPT-5.6 Sol, released last month, but with a specialized focus on advanced security research tasks, vulnerability development, and attack testing. The company highlighted that this model represents a significant leap in offensive and defensive capabilities within the cybersecurity field. According to the Artificial Intelligence Laboratory on its website, the new model was subjected to benchmarks such as ExploitGym, designed to evaluate whether agents can convert vulnerabilities into functional exploits. In the results, GPT-5.6-Cyber outperformed both GPT-5.6 Sol and GPT-5.5-Cyber, demonstrating superior performance in exploit generation. One area where OpenAI strengthened GPT-5.6-Cyber is its ability to accurately identify and calibrate the severity of zero-day vulnerabilities, i.e., security flaws unknown to developers and without a fix available at the time of discovery. This capability is critical for anticipating threats before they are exploited. Compared to GPT-5.6 Sol, GPT-5.6-Cyber completes 95 percent of advanced cybersecurity and dual-use requests, while Sol rejects almost all, achieving only about 1.5 percent completion. This difference underscores the new model’s specialized focus. As a concrete example, OpenAI researchers stated that they used GPT-5.6-Cyber to identify previously unknown vulnerabilities in V8, Chrome’s JavaScript engine, which could be chained to corrupt memory and escape the heap sandbox. This demonstrates its usefulness in real-world high-risk scenarios. When it launched its frontier model GPT-5.6 Sol, the tech company included a series of cybersecurity safeguards. These restrictions are now lifted with the Daybreak program, which expands its availability to more organizations, enabling broader use of these advanced tools. The company has split Daybreak access into two modalities with different verification requirements. On one hand, Daybreak Blue is an adjusted level for the usual defensive work that companies or organizations may require, such as code review or malware analysis. This level provides access to general models like GPT-5.6 Sol, removing the restrictions that apply to models available to all users. On the other hand, Daybreak Red offers advanced access to cutting-edge models for researchers and testing teams, and includes the new specialized GPT-5.6-Cyber. This level is designed for offensive cybersecurity tasks and in-depth vulnerability analysis. OpenAI clarified that access to these levels is limited to “approved individuals and organizations” conducting authorized work, with identity verification, account security, monitoring, approved usage restrictions, and legal statements. This ensures strict control over who can use these powerful tools. The company is expanding Daybreak through security and services partners, including Accenture, Cisco, Cloudflare, CrowdStrike, IBM, Palo Alto Networks’ Unit 42, and others. OpenAI noted that this allows more organizations to have a pathway to use cutting-edge cybersecurity models through trusted products, managed services, and client collaborations. “This enables more organizations to have a pathway to use cutting-edge cybersecurity models through trusted products, managed services, and client collaborations.” Through the program, OpenAI indicated that community initiatives are created, referring to alliances with security companies as well as with the open-source community, and AI-assisted review is promoted to find and fix flaws. In this way, the company aims to democratize access to state-of-the-art cybersecurity tools. Source: Infobae

OpenAI has expanded on August 10 with OpenAI Daybreak and introduced GPT-5.6-Cyber, a model specialized for advanced and authorized cybersecurity research. The company now splits access into two levels, Blue and Red, and extends the program to security and service providers who can incorporate these capabilities into products, managed services, and client projects. The move shifts part of the debate from the model’s pure capability to who can use it, under what controls, and within what operating environment. GPT-5.6-Cyber reduces rejections in advanced work The most visible difference appears in requests that generalist models typically block due to dual-use concerns. OpenAI has created an internal evaluation, Advanced Cybersecurity Completion Rate, to measure the proportion of advanced requests the system answers. GPT-5.6-Cyber completes 95.0%, compared with 1.5% for GPT-5.6 Sol with the usual safeguards and 2.0% for Sol under Daybreak Blue. GPT-5.5-Cyber reaches 57.3%. Daybreak Blue allows using frontier generalist models, including GPT-5.6 Sol, with safeguards tailored to authorized defensive work and fewer system-level blocks. Sol still maintains refusals for certain dual-use requests. Daybreak Red adds models trained specifically for vulnerability research, exploit validation, and security testing, with GPT-5.6-Cyber as the new reference. Greater ability to respond does not imply uniform superiority. In ExploitGym, focused on turning known vulnerabilities into functional exploits within controlled environments, GPT-5.6-Cyber surpasses GPT-5.6 Sol and GPT-5.5-Cyber. It also achieves better results than Sol in an internal zero-day vulnerability discovery evaluation. The pattern shifts in Vulnerability Discovery and Report Writing: GPT-5.6 Sol scores better, a difference OpenAI attributes to specialized-model reports that are sometimes shorter and less detailed. Another contrast appears in ExploitBench, a test of exploiting V8 vulnerabilities with defensive protections activated. With a standard 300-turn limit, GPT-5.6 Sol under Daybreak Blue completes tasks with greater token efficiency and achieves the best result. When the limit is extended to 600 turns, the gap with GPT-5.6-Cyber narrows. For security teams, specialization broadens coverage of sensitive tasks, while the generalist retains advantages in certain tasks. Daybreak separates access by type of work OpenAI recommends Daybreak Blue as the entry point for most defenders. Its use cases include vulnerability discovery, safe code review, malware analysis, incident response, and patch validation. Daybreak Red is reserved for more specialized authorized research, including exploit development and validation, pentesting, and red teaming. Access is not open. The company requires prior approval and implements identity verification, account security, monitoring, usage restrictions, and legal declarations. As of September 1, 2026, individual Daybreak accounts will also be required to use physical security keys. OpenAI is also guiding Codex users toward the “auto-review” mode, which reviews actions requiring elevated permissions and may block operations with destructive risk. Operational recommendations include isolating flows in sandboxes, limiting access to production systems or the Internet, monitoring agent actions, and precisely defining authorized permissions. Thus, reducing safeguards for certain users is accompanied by more controls around identity, execution environment, and the scope of work. GPT-5.6-Cyber brings research to V8 The most relevant part outside benchmarks comes from V8, the JavaScript engine used by Chrome. OpenAI researchers used GPT-5.6-Cyber to locate two previously unknown vulnerabilities that, when combined, could allow memory corruption and an escape from V8’s heap sandbox. The findings were disclosed to Google through coordinated disclosure. One was fixed and has the identifier CVE-2026-15903. The flaw affected the V8 optimizer compiler and could lead to bypassing a bounds check in certain value-conversion scenarios, with the possibility of reading or overwriting memory within the isolated environment. The second vulnerability remains, according to OpenAI, within the coordinated disclosure process. The company also states that with GPT-5.6-Cyber it identified at least five vulnerabilities in a mobile operating system, three critical vulnerabilities in a popular database, and more than 400 issues capable of facilitating privilege escalation in the kernel of an operating system. The affected products have not been publicly identified as the coordination to fix them continues. Partners turn Daybreak into a business channel Daybreak’s expansion also rests on consultancies, integrators, and security manufacturers. The Daybreak Cyber Partner Program includes Accenture, IBM, Capgemini, EY, PwC, and NCC Group, alongside Unit 42 of Palo Alto Networks, CrowdStrike, Cisco, Sophos, Akamai, Fortinet, and Cloudflare. The distribution model has a peculiarity: access to the underlying models remains in the hands of the approved partner and is not directly transferred to the client. Organizations can receive Daybreak capabilities within products, managed services, or projects governed by those providers. For companies that do not want to build their own AI-for-cybersecurity program, the model reduces part of the technical burden while adding dependence on the provider’s controls and processes. The change also expands the role of professional services. Detecting a vulnerability is only the first step; protection depends on validating the risk, identifying affected systems, developing a fix, and deploying it to production. OpenAI frames that remediation cycle as a bottleneck: a larger volume of AI-assisted findings could increase the workload if teams cannot verify and fix as quickly. Government gains weight alongside performance OpenAI classifies both GPT-5.6 Sol and GPT-5.6-Cyber at the “High” level of cybersecurity capability in its Preparedness Framework, below “Critical.” The company has not yet published the specific system card for GPT-5.6-Cyber and notes that it will publish additional evaluations later. Some results from the launch come from internal benchmarks, a relevant limit when comparing percentages across models and scenarios. The expansion arrives three days after OpenAI announced that, based on preliminary evaluations of a future model named Astra, it could no longer rule out “Critical” cybersecurity capabilities. The company announced stricter internal controls, isolated environments, network and tool restrictions, additional monitoring, and testing with public bodies and AI security entities. For companies, Daybreak turns that technical evolution into an architectural and governance decision. Teams will need to determine which tasks remain in generalist models, which justify Red access, and which part is best delegated to a specialized provider. Reducing rejections can accelerate research, but it also raises the demands for identity, isolation, monitoring, and permission delimitation. The model’s performance becomes another piece of the surrounding security system.

OpenAI has expanded its Daybreak cyber defense service to include two tiers: Blue and Red. Blue offers basic cybersecurity tools like incident response and malware analysis, while Red provides advanced capabilities including the new GPT-5.6-Cyber model designed for specialized security testing and vulnerability research. This new model is currently available only to trusted partners such as Accenture, IBM, and Crowdstrike. The expansion addresses rising AI-driven cyber threats by equipping defenders with cutting-edge AI tools to counter increasingly sophisticated attacks. US Stocks, ETFs, and Options are facilitated by PT PG Berjangka as a Derivatives Broker which is licensed and supervised by the Financial Services Authority of Indonesia (OJK) for financial derivative products with underlying securities. Transactions are recorded on the Jakarta Futures Exchange (JFX) and the Indonesian Futures Clearing House (KBI). Indonesian Stocks (by PT Sarana Santosa Sejati as a Level-II Institutional Security Brokers Marketing Partner, in collaboration with PT Pluang Maju Sekuritas as a Securities Company) are licensed and supervised by the Financial Services Authority of Indonesia (OJK). Crypto Assets and Digital Financial Asset Derivatives (Crypto Futures) are facilitated by PT Bumi Santosa Cemerlang which is licensed and supervised by the Financial Services Authority of Indonesia (OJK). Transactions are recorded by Central Finansial X (CFX) and are guaranteed by Kliring Komoditi Indonesia (KKI). Mutual Funds are facilitated by PT Sarana Santosa Sejati as a Mutual Fund Securities Selling Agent (APERD) which is licensed and supervised by the Financial Services Authority of Indonesia (OJK). Gold is facilitated by PT Pluang Emas Sejahtera as a Digital Gold Market Maker, which is licensed and supervised by the Commodity Futures Trading Regulatory Agency (Bappebti). All Gold is stored by PT ICDX Logistik Berikat (ILB) in collaboration with PT Brinks Solutions Indonesia (Brink’s), recorded by the Indonesia Commodity & Derivatives Exchange (ICDX), and is guaranteed by the Indonesian Clearing House (ICH).

Every day seems to brings fresh news of an AI agent going “rogue.” Whether that’s compromising Hugging Face, hacking a gym website, or creating its own fake profiles to socially engineer an intrusion, AI models are increasingly behaving like bad actors. So, the AI labs that make the models doing the hacking are expanding their cyber protection offerings. This week, OpenAI announced an expansion of Daybreak, its cyber defense service which it launched earlier this year, not long after Anthropic released its cyber-focused model Mythos. Daybreak is a service that bundles access to models, tools and workflows for defenders. The expansion includes access to a brand new cyber-focused model designed for defensive work. OpenAI said Monday that Daybreak would now consist of two tiers: Blue and Red. Both of these tiers will allow approved customers access to OpenAI’s limited-access frontier cyber models. Frontier models — the most advanced available — have been a subject of controversy. The Trump administration previously sought to collaborate with AI companies on the roll out of such models, purportedly over safety concerns. Previously, OpenAI deployed significant guardrails to using these models, limiting what customers could do with them. Blue, which appears to be the more basic of the two, offers a variety of cyber services, including incident response, malware analysis, and patch validation. OpenAI calls Blue its “recommended starting point for most defenders,” implying that it should be more than enough for most enterprises. Red, on the other hand, offers a broader and potentially more dangerous toolkit. The company grants its users “purpose-trained cybersecurity models,” designed to carry out security testing and vulnerability research. With Red also comes the new model, GPT‑5.6‑Cyber, which is only available at that tier. 5.6-Cyber is built off of GPT‑5.6 Sol, and offers enhanced capabilities for certain specialized cybersecurity tasks, the company said. At the moment, GPT‑5.6‑Cyber is only being made available for “trusted customer partners,” including reportedly Accenture, IBM, Crowdstrike, Cloudflare, and others. While the threats from AI agents are rapidly increasing, critics have also pointed out that they function as marketing opportunities for the AI labs. OpenAI is certainly marketing its upgraded Daybreak that way. “The cybersecurity world is rapidly changing—threat actors will increasingly use AI to conduct cyberattacks at unprecedented speed and scale, including in fully autonomous ways,” the company said in a blog post. “As these capabilities spread, defenders have a narrowing window to prepare.” At the same time, enterprises remain interested in buying their protection from the AI labs who know the security risks best, because they know them first-hand.

OpenAI is doubling down on cybersecurity defense as AI agents become both tools and targets in an escalating digital arms race. The company's expanding its Daybreak initiative - launched just three months ago - to give security partners early access to its most powerful models for threat detection. The move signals growing concern that AI systems themselves are creating entirely new attack surfaces that traditional security tools can't handle. OpenAI is scaling up its cybersecurity efforts at a moment when AI systems are fundamentally changing how attacks happen. The company's Daybreak initiative, quietly introduced back in May, is getting a significant expansion according to reports from CNBC. The timing isn't coincidental - as AI agents become more autonomous, they're creating security challenges that didn't exist even a year ago. Daybreak works by giving select ecosystem partners access to OpenAI's most advanced models, specifically adapted for cybersecurity work. These aren't general-purpose chatbots - they're systems trained to spot anomalies, predict attack patterns, and respond to threats in real-time. The idea is that the same technology powering AI agents can be turned around to defend against malicious ones. The threat landscape is evolving faster than most companies can track. AI-powered phishing attempts now adapt their language in real-time based on victim responses. Automated systems probe networks for vulnerabilities at machine speed. And as more businesses deploy AI agents with access to sensitive systems, each one becomes a potential entry point. Traditional signature-based security tools struggle because these attacks don't follow predictable patterns. What makes Daybreak different is the partnership model. Rather than building its own security products, OpenAI is essentially arming established cybersecurity firms with AI capabilities they couldn't develop alone. Partners get early access to model updates, specialized training data, and direct collaboration with OpenAI's safety team. It's a recognition that the company can't solve enterprise security problems by itself. The expansion comes as competitors are making similar moves. Microsoft has integrated AI-powered threat detection across its enterprise stack. Google is using its models to analyze security telemetry at scale. But OpenAI's approach is more platform-like - providing the underlying intelligence that other companies build defenses around. There's an interesting tension here. The same technology that makes AI agents useful - their ability to understand context, chain together actions, and operate with minimal oversight - also makes them dangerous in the wrong hands. A compromised AI agent with access to company systems could exfiltrate data, manipulate records, or sabotage operations while appearing to function normally. Defenders need AI that can think like an attacker. The cybersecurity industry has struggled with AI hype for years, with vendors slapping "AI-powered" labels on conventional tools. What's different now is that the threats themselves are genuinely AI-native. You can't defend against an adaptive, learning attack system with static rules. The defensive tools need to evolve as fast as the offensive ones. For enterprises deploying AI agents, this creates a new calculus. The productivity gains are real - agents can handle customer service, process documents, manage workflows. But each deployment expands the attack surface. Security teams that were already stretched thin now need to monitor AI behavior, validate agent actions, and detect when systems have been compromised or manipulated. OpenAI hasn't disclosed which partners are part of Daybreak or how many companies have joined since May. That opacity is typical for security programs, where publicizing defensive capabilities can help attackers work around them. But the expansion itself suggests the initial cohort found value - and that OpenAI sees cybersecurity as a growth area, not just a responsibility. The broader question is whether AI defenses can actually keep pace with AI attacks, or if we're heading toward an escalating arms race that favors offense. History suggests attackers usually have the advantage - they only need to find one weakness, while defenders have to protect everything. Adding AI to both sides might just speed up the cycle without fundamentally changing the dynamics. What's clear is that cybersecurity can no longer be an afterthought in AI deployment. Companies rushing to adopt agent systems need to think about security from the start, not bolt it on later. And that means working with partners who understand both the technology and the threat landscape - exactly what Daybreak is positioning itself to provide. OpenAI's Daybreak expansion is less about a single product launch and more about acknowledging a fundamental shift in how security works when AI agents are everywhere. The companies that figure out AI-powered defense early will have a significant advantage as enterprises scale up agent deployments. Those that don't will find themselves defending against machine-speed attacks with human-speed tools - a fight they can't win. The real test will be whether collaborative programs like Daybreak can actually close the gap, or if we're just entering a new phase of the eternal cat-and-mouse game between attackers and defenders.

OpenAI is splitting its cybersecurity program into two access tiers and releasing a new purpose-trained model alongside them, the company announced on August 10, 2026. Daybreak Blue opens frontier general-purpose models, including GPT-5.6 Sol, to approved defenders for everyday security work, while Daybreak Red gates the new GPT-5.6-Cyber model behind tighter vetting for vulnerability research, exploit validation, and security testing. The structure addresses a tension OpenAI says it has been managing in production. The system-level safeguards it runs on GPT-5.6 Sol screen cybersecurity-related requests to prevent misuse, but the company says those same screens block legitimate defensive work. Daybreak Blue removes them for verified users. A residue of highly dual-use prompts, such as penetration testing production systems, still draws refusals even under Blue access, which is where GPT-5.6-Cyber comes in: a version of GPT-5.6 Sol trained to reduce refusals on advanced cybersecurity tasks and to improve on specialized work like finding zero-day vulnerabilities and developing exploit chains. What the evaluations measured To quantify how much more permissive the new model is, OpenAI built an internal evaluation it calls the Advanced Cybersecurity Completion Rate, measuring how often models respond to requests involving exploit-chain development, authentication bypass, privilege escalation, and similar scenarios. On that benchmark, the company reports, GPT-5.6-Cyber completes 95.0% of requests, against 1.5% for GPT-5.6 Sol under standard safeguards and 2.0% for GPT-5.6 Sol under Daybreak Blue. The prior purpose-trained model, GPT-5.5-Cyber, completed 57.3%, a refusal rate the company says generated persistent complaints from security researchers. Capability evaluations tell a more qualified story. On ExploitGym, which tests whether agents can turn known vulnerabilities into working exploits that achieve code execution in controlled environments, OpenAI says GPT-5.6-Cyber outperforms both GPT-5.6 Sol and GPT-5.5-Cyber. On the company’s internal Vulnerability Discovery and Report Writing evaluation, GPT-5.6-Cyber improves over GPT-5.5-Cyber but lands below GPT-5.6 Sol, a result OpenAI attributes to the model producing shorter, less detailed reports. On ExploitBench, a harder exploitation task with the V8 sandbox enabled and less information given to the agent, the general-purpose GPT-5.6 Sol performs best within the standard 300-turn limit, with the gap narrowing when runs extend to 600 turns. All of these figures are vendor-reported, several on internal benchmarks that outside evaluators have not replicated. The claims with the most weight behind them are not benchmarks at all. OpenAI says it used GPT-5.6-Cyber to investigate V8, the JavaScript engine inside Chrome, and uncovered two previously unknown vulnerabilities that could be chained to corrupt memory and escape the V8 heap sandbox. The first, a compiler bug in which a skipped safety check lets an attacker read or overwrite memory inside Chrome’s sandbox, was reported to Google through coordinated disclosure, fixed, and assigned CVE-2026-15903. The company also lists, without naming the affected software, at least five vulnerabilities in a popular mobile operating system including a privilege-escalation chain from an untrusted app, three critical vulnerabilities in a popular database including a remote path to code execution, and more than 400 privilege-escalation vulnerabilities in a popular operating system kernel, all moving through disclosure with Daybreak partners and open-source maintainers. SpecterOps, the security firm whose CTO Jared Atkinson tested the model early, described the results in terms of work compression: the model “has completed work in under a day that earlier models had not resolved after weeks of intermittent effort.” How the two tiers are governed Access to both tiers runs through identity verification, account security requirements, monitoring, approved-use restrictions, and legal attestations, with separate application paths for individuals and organizations. OpenAI is pushing Daybreak customers using its Codex coding agent from full-access mode toward an auto-review mode that evaluates actions requiring elevated permissions before execution, and the company will require hardware security keys on all individual Daybreak accounts beginning September 1, 2026. A system card with further evaluations of GPT-5.6-Cyber is planned for a later date. Under OpenAI’s Preparedness Framework, both GPT-5.6 Sol and GPT-5.6-Cyber were assessed as High for cybersecurity capability and below the Critical threshold. That assessment lands days after Unite.AI reported that OpenAI’s upcoming Astra model may cross the Critical cybersecurity threshold, and the company used the announcement to reiterate a point from its earlier incident disclosures: GPT-5.6-Cyber was not involved in the exploitation of Hugging Face, and no model with that involvement is slated for release. Where Daybreak stood before this release The tiered structure consolidates a program that had been expanding in pieces. OpenAI launched the full version of GPT-5.5-Cyber on June 22, 2026 alongside a Daybreak Cyber Partner Program counting Accenture, CrowdStrike, Cisco, IBM, and Palo Alto Networks among its participants, and Patch the Planet, an open-source remediation initiative founded with Trail of Bits. On that earlier release, OpenAI reported GPT-5.5-Cyber reaching 85.6% on CyberGym against 81.8% for GPT-5.5, and 39.5% against 25.95% on ExploitGym. Per the June 22 post, more than 30 open-source projects committed to participate, and the initial five-day sprint surfaced hundreds of issues with dozens of patches merged. Daybreak Blue and Red replace what had been a single Trusted Access track with a two-rung ladder: the general-purpose frontier, guardrails relaxed, for the broad defender base, and a refusal-light specialist model for the smaller group whose authorized work runs to exploit development and red teaming.

OpenAI has launched GPT-5.6-Cyber, a cybersecurity-focused model capable of completing 95% of advanced exploitation requests that the public layer rejects, within a verified and reviewed access tier. Key points: - GPT-5.6-Cyber responds to 95% of advanced cybersecurity requests, compared to only 1.5% for GPT-5.6 Sol. - The Daybreak platform now operates on two levels: Blue for defense and Red for offensive research. - The model discovered two vulnerabilities in the Chrome engine and Google fixed them under CVE-2026-15903. GPT-5.6-Cyber restructures Daybreak into Blue and Red. The company announced the model on August 10 and reorganized the Daybreak defense program around two separate access levels. Daybreak Blue provides authorized defenders with a version of GPT-5.6 Sol with strict system-level restrictions removed for tasks such as malware analysis, incident response, and validation of security patches. Daybreak Red goes a step further, with models trained specifically for legally authorized vulnerability research, exploitability verification, and penetration testing and security. In an internal standard that included developing exploit chains, bypassing authentication systems, and privilege escalation, the new model answered 95% of requests, compared with 1.5% for GPT-5.6 Sol and 57.3% for the previous GPT-5.5-Cyber. When running the same model over the Blue access level, its responses drop to 2% only, reflecting a clear difference in exploitation capabilities between the two tiers. Researchers directed the model toward V8, the JavaScript engine used by Google Chrome, to uncover two previously unknown vulnerabilities that could be combined to cause memory corruption. Google addressed the issue after disclosure by a coordinator and logged it under CVE-2026-15903 as a high-severity browser engine vulnerability. The model also extracted more than 400 vulnerabilities for privilege escalation in a widely used operating system kernel. Related article: OpenAI buys back employee shares for $7 billion... and the IPO still has no date. Security researchers weigh rejection rates against abuse risks The defenders have complained for months about rising rejection rates in high-capacity models, a tension closely watched by the sector as labs strive to keep capabilities away from attackers. The gap has been tracked by the industry through various reports, including one from Axios. OpenAI filters applicants through identity verification, usage monitoring mechanisms, limits on permitted uses, and legal commitments, while major companies such as Accenture, IBM, CrowdStrike, and Cisco are able to integrate these models into their commercial security products. Jared Atkinson, CTO at SpecterOps, said the model assesses real-world exploitation constraints far more accurately than its predecessors and credited it with closing research projects in under a day after they had stalled for weeks with earlier systems. OpenAI classified GPT-5.6-Cyber as “High” in cybersecurity capabilities under the Preparedness Framework, one level below “Critical.” Incidents of “Unchecked Intelligence” shape OpenAI’s cybersecurity policy The launch followed a series of test incidents in which the models reached systems that should have remained outside their scope. OpenAI agents exited isolated sandbox environments and breached Hugging Face, while Anthropic reported that Claude models reached three institutions, and Meta confirmed a breach at a third-party company. In the past week, OpenAI tightened its rules governing the use of its most capable models. On August 7, it announced a slowdown in the development of Astra, an unconstrained model that it said it could not rule out possessing cyber capabilities classified as critical. The company emphasizes that GPT-5.6-Cyber played no role in the Hugging Face breach, while physical security keys will become mandatory for all Daybreak accounts starting September 1. Also see: The foldable iPhone has not launched yet… but Apple planned two additional versions through 2028.

OpenAI continues expanding Daybreak, its initiative focused on strengthening the capabilities of its models for advanced cybersecurity tasks. The company has published new results for GPT-5.6-Cyber, a model trained specifically for certain workflows related to security research and controlled exploit development, with the aim of responding to a scenario in which the window to detect and fix vulnerabilities is increasingly narrow. One of the areas where OpenAI has evaluated the model is ExploitGym2, a test that analyzes whether an agent can convert known vulnerabilities into functional exploits capable of achieving arbitrary code execution within controlled environments. In this evaluation, GPT-5.6-Cyber outperforms GPT-5.6 Sol and GPT-5.5 Cyber, according to results published by OpenAI. The model has also been trained to improve its ability to identify zero-day vulnerabilities and correctly determine their severity. To this end, OpenAI created an internal evaluation in which it provides the models with an up-to-date version of an open-source code repository and asks them to investigate possible vulnerabilities, develop high-impact proof-of-concept tests, and prepare a technical report. The models are evaluated both for the severity and impact of discoveries and for the accuracy with which they calibrate their importance and the quality of the reports generated. In this test, GPT-5.6-Cyber surpassed GPT-5.6 Sol, something OpenAI attributes to its specialized training. The results are not uniform across all tests; in the internal Vulnerability Discovery and Report Writing evaluation, the models receive an open-ended task to locate vulnerabilities within a repository that contains a known vulnerability. To obtain a good score they must find severe and actionable vulnerabilities, develop a functional proof-of-concept, and present a detailed security report. Both GPT-5.6 Sol and GPT-5.6-Cyber show improvement over GPT-5.5-Cyber, yet in this test GPT-5.6-Cyber achieves worse results than GPT-5.6 Sol. OpenAI suggests one reason is that the specialized model sometimes generates shorter and less detailed vulnerability reports. It also used ExploitBench3, an evaluation focused on developing a complete exploit from a vulnerability in V8, the JavaScript engine used by Chrome. The test is more complex than ExploitGym because it maintains more defensive protections active, including V8’s sandbox, and provides the agent with less information about the vulnerability. In the standard scenario, limited to 300 turns, GPT-5.6 Sol is the best-performing model and also manages to solve the tasks using fewer tokens. However, when the limit is extended to 600 turns, the performance gap between GPT-5.6 Sol and GPT-5.6-Cyber narrows. Beyond benchmarks, OpenAI asserts that GPT-5.6-Cyber has also been used in vulnerability research on real software. One example provided by the company concerns V8, the JavaScript engine used by Chrome. During this investigation, GPT-5.6-Cyber helped discover two previously unknown vulnerabilities that could be chained to cause memory corruption and escape V8’s heap sandbox. OpenAI researchers validated the discoveries and reported them to Google through a coordinated vulnerability disclosure process. Google subsequently fixed the issue and assigned it the CVE-2026-15903 identifier. According to OpenAI, CVE-2026-15903 is a high-severity vulnerability related to the V8 compiler optimizer, involving a value-to-integer conversion in which the compiler improperly skipped a security check. This could cause certain undefined values to generate an unexpectedly large number, and if that number ended up being used as an array index, the compiler could wrongly assume the index was in bounds and skip the corresponding check. The result could allow an attacker to read or overwrite memory belonging to other objects, potentially opening the door to arbitrary code execution within Chrome’s environment. Therefore, to escape the heap sandbox, it was necessary to combine this vulnerability with a second flaw. OpenAI states that GPT-5.6-Cyber also managed to find that second vulnerability, enabling the chaining of both issues. The company also provided early access to GPT-5.6-Cyber to a group of trusted clients to test its usefulness in real defensive workflows. Jared Atkinson, CTO of SpecterOps, notes that the model has improved vulnerability research by reasoning more precisely about exploit constraints and by maintaining better context of complex investigations. According to OpenAI, these experiences show that cybersecurity-specialized models can complement human researchers in processes that require analyzing large amounts of code, forming hypotheses, testing them, and repeating the process until an exploitable vulnerability is found. This is the opinion of internet users, not Zonamovilidad.es. It is not permitted to post comments that contravene the law, are defamatory, or contain insults, personal attacks, or opinions that undermine the dignity of others. Tone-deaf or unclear messages will not be accepted either. The email address requested will not be used for commercial purposes. Your email address will not be published. We reserve the right to delete comments that we consider off-topic.

OpenAI launches GPT-5.6-Cyber to combat cyberattacks We bring you our visitors the most important and latest news in the following article: OpenAI launches GPT-5.6-Cyber to combat cyberattacks, today Tuesday, August 11, 2026, 12:26 AM. OpenAI (OpenAI) announced the launch of the GPT-5.6-Cyber model, a specialized edition designed to help researchers and cybersecurity teams detect, test, and respond to advanced digital threats, in a new step to enhance the use of artificial intelligence in the field of cybersecurity. The new model’s release comes as Daybreak is expanded, a program through which the company provides its models and specialized tools to accredited security authorities, at a time when concerns about using AI to carry out faster and more autonomous cyberattacks are increasing. GPT-5.6-Cyber is built on GPT-5.6 Sol, but underwent specialized training focusing on a number of advanced cybersecurity tasks, including discovering new vulnerabilities, analyzing software, and developing and verifying exploit chains. OpenAI aims with this model to address one of the problems security researchers face when using general AI models, namely the refusal to execute certain cybersecurity-related requests, even when their purpose is defensive and legitimate. The company says GPT-5.6-Cyber is designed to reduce these refusals in some high-risk, dual-use scenarios, with access restricted to entities that have received OpenAI approval. OpenAI disclosed internal test results measuring the models’ ability to handle advanced cybersecurity requests. The test showed that GPT-5.6-Cyber responded to 95% of requests involving scenarios such as exploit-chain development, bypassing authentication systems, and privilege escalation. In contrast, GPT-5.6 Sol recorded a 1.5% response rate when run with standard safeguards, rising to 2% when the model was used through Daybreak Blue. GPT-5.6-Cyber also outperformed GPT-5.5-Cyber, which had about a 57.3% response rate for the same requests. OpenAI also announced simultaneously the expansion of the Daybreak program to include two different access levels, aligned with the nature of tasks performed by security entities. This level provides accredited defenders with access to advanced general AI models, including GPT-5.6 Sol, with modifications to security barriers designed to protect cybersecurity uses. OpenAI recommends using Daybreak Blue as the starting point for most security teams, especially for the following tasks: - Vulnerability discovery - Code review - Malware analysis - Incident response - Vulnerability management - Testing security updates and fixes - Assessing system protection Daybreak Red targets the most advanced security tasks, as it allows accredited entities access to specialized models such as GPT-5.6-Cyber. This level is dedicated to tasks involving advanced vulnerability research, exploit verification, penetration testing, and red-team activities. Cybersecurity companies face increasing challenges as AI models evolve; restrictions intended to prevent misuse may also prevent researchers from carrying out necessary defensive operations. OpenAI notes that AI-powered cyber threats could evolve rapidly in the coming period, including attacks capable of operating almost autonomously or completely autonomously. From this, the company believes that giving trusted defenders access to more capable models can help them detect and close vulnerabilities before attackers exploit them. Among the most notable examples OpenAI disclosed was the use of the new model in analyzing V8, the JavaScript engine underlying Google Chrome. During security testing, GPT-5.6-Cyber helped discover two previously unknown vulnerabilities that could be exploited within a chain leading to memory corruption and bypassing some isolation mechanisms in V8. OpenAI researchers validated the results and reported them to Google following a coordinated disclosure process, before patching the vulnerability which received CVE-2026-15903. The vulnerability relates to an issue in the V8 optimization translator, which could bypass a security check when converting certain values to integers, potentially allowing access to unauthorized memory locations under certain conditions. GPT-5.6-Cyber results were not limited to the V8 engine, as OpenAI said the model helped discover high-severity vulnerabilities across a range of software and systems. Among the disclosed results were: - At least five vulnerabilities in a mobile operating system, including a chain that could lead from an untrusted app to privilege escalation. - Three critical flaws in a well-known database, including one that enables a remote code execution path. - More than 400 vulnerabilities that could lead to privilege escalation in a widely used kernel. OpenAI confirmed it is working with partners and open-source software developers to disclose these security issues and fix them. Despite the model’s significant potential, OpenAI states that GPT-5.6-Cyber did not exceed a high level in cybersecurity capability ratings within the company’s readiness framework. According to OpenAI, the new model, like GPT-5.6 Sol, remains below the critical threshold that marks the highest level of cyber capability within the readiness framework. The company noted that GPT-5.6-Cyber surpassed GPT-5.6 Sol in some tasks it was trained for in a specialized manner, but the improvement was not enough to place it in the critical capability category. The announcement of the new model came as OpenAI continued investigating an incident related to the Hugging Face platform being hacked using AI tools. Two OpenAI employees revealed at the Black Hat cybersecurity conference that agents created a message board to exchange information about vulnerabilities they had found, which later aided the breach. However, OpenAI stressed that GPT-5.6-Cyber did not participate in hacking Hugging Face, and the model is not part of the breach that was previously disclosed. GPT-5.6-Cyber will not be available to all users, as OpenAI requires credentialing processes before granting institutions or individuals the right to use advanced models within Daybreak. Security procedures include identity verification, account protection, activity monitoring, adherence to usage policies, and legal requirements specific to participants. The company also encourages Codex participants in the program to use Auto-Review mode, which allows reviewing high-privilege actions before execution. OpenAI also announced that individual Daybreak participants will be required to use hardware security keys starting September 1, 2026.