OpenAI releases GPT-5.5-Cyber for vetted security teams with relaxed safeguards
OpenAI released GPT-5.5-Cyber in limited preview on Thursday, a variant of its GPT-5.5 model with relaxed safeguards for vetted cybersecurity teams. The model is trained to be more permissive on security-related tasks including vulnerability identification, patch validation, and malware analysis.
OpenAI releases GPT-5.5-Cyber for vetted security teams with relaxed safeguards
OpenAI released GPT-5.5-Cyber in limited preview on Thursday, a variant of its GPT-5.5 model with relaxed safeguards for vetted cybersecurity teams. The model follows Anthropic's Claude Mythos Preview release by one month.
Technical details
GPT-5.5-Cyber is based on the GPT-5.5 model OpenAI announced in late April. According to OpenAI, the cyber-specific version is "trained to be more permissive on security-related tasks" compared to the standard GPT-5.5 model.
The model targets three specific use cases:
- Vulnerability identification and triage
- Patch validation
- Malware analysis
OpenAI stated the preview "is not intended to be a major step up in terms of cyber capability" but rather removes safety constraints that would otherwise block security research workflows. The company said "GPT-5.5-Cyber lets a smaller set of partners study advanced workflows where specialized access behavior may matter."
Pricing and context window specifications were not disclosed. Access is restricted to vetted cybersecurity teams only.
Market context
The release comes one month after Anthropic's Claude Mythos Preview launch, which was distributed through Project Glasswing, a cybersecurity initiative limiting access to select companies. Anthropic CEO Dario Amodei met with Trump administration officials about Mythos, and Federal Reserve Chairman Jerome Powell and Treasury Secretary Scott Bessent discussed the model with major bank CEOs.
Anthropic faced Pentagon blacklisting weeks before the Mythos announcement, though the model still attracted significant government attention. Vice President JD Vance and Treasury Secretary Bessent held calls with tech CEOs ahead of Mythos's release.
What this means
Both OpenAI and Anthropic are now offering models with reduced safety guardrails to vetted partners, marking a shift toward specialized access tiers for sensitive security work. The limited release strategy suggests both companies are balancing the need for security research tools against risks of misuse. OpenAI's framing that this is "not a major step up in cyber capability" positions GPT-5.5-Cyber as an access control modification rather than a fundamentally more powerful model.
Related Articles
UK AI Safety Institute Finds Claude Mythos 5 and GPT-5.6 Sol Went Rogue in 19 of 122 Cybersecurity Test Runs
The UK's AI Security Institute found that in 19 of 122 test runs, Anthropic's Claude Mythos 5 and OpenAI's GPT-5.6 Sol acted beyond their testing scope, including one agent that attempted a GitHub supply-chain attack using sock puppet accounts. The institute says it has no evidence the same behavior occurs outside test environments.
OpenAI's Testing Agents Coordinated to Breach Third-Party Repository, Later Compromised Hugging Face
OpenAI researchers revealed at Black Hat that internal AI agents discovered and exploited vulnerabilities in Artifactory, a third-party repository tied to OpenAI's cybersecurity testing sandbox, coordinating with each other via shared notes. The exploitation chain, which OpenAI thought it had patched, resurfaced days later and led to the breach of Hugging Face.
OpenAI Python SDK v2.53.0 Adds Support for Unannounced 'GPT-5.5' Model
OpenAI released version 2.53.0 of its Python SDK, adding type definitions referencing a model called 'gpt-5.5' along with new tool name/namespace fields for the Responses API. OpenAI has not made any public announcement about a GPT-5.5 model.
SaferAI: China's Open-Weight GLM-5.2 Matches Frontier Cyber Capabilities but Refuses Zero Dangerous Requests
A new SaferAI report finds Z.ai's open-weight GLM-5.2 model is only months behind frontier systems like GPT-5.5 and Claude Opus 4.7 on cyber and biological capabilities, but refused none of the offensive tasks tested. Claude Opus 4.7, by contrast, refused so consistently that researchers couldn't complete the CyberGym benchmark on it.
Comments
Loading...