OpenAI restricts cybersecurity AI access following Anthropic's model controls
OpenAI is restricting access to a new AI model with advanced cybersecurity capabilities to a small group of companies, mirroring Anthropic's decision to limit distribution of its Mythos Preview model. OpenAI's move builds on its February launch of the Trusted Access for Cyber pilot program following GPT-5.3-Codex, offering $10 million in API credits to participants.
OpenAI Restricts Cybersecurity AI Access Following Anthropic's Model Controls
OpenAI is limiting access to a new AI model with advanced cybersecurity capabilities to a select group of companies, according to Axios reporting. The decision directly mirrors Anthropic's approach announced this week when it restricted access to Mythos Preview to only tech and security firms.
Current Programs and Models
OpenAI launched its "Trusted Access for Cyber" pilot program in February following the release of GPT-5.3-Codex, described as the company's most capable cybersecurity model to date. The pilot provides participants with access to particularly powerful models for defensive security work, backed by $10 million in API credits.
Anthropically's Mythos Preview model represents the company's latest cybersecurity-focused offering. The company has explicitly ruled out a public release, stating that models in the Mythos class will not ship to the broader market until adequate safety guardrails are in place.
Industry-Wide Safety Posture
Both companies cite the powerful hacking capabilities embedded in these models as the primary reason for restricting distribution. Rather than making advanced cybersecurity capabilities universally available, the companies are implementing staged rollouts to trusted partners in security and technology sectors.
Anthropic's decision to permanently exclude public access represents a stricter stance than OpenAI's approach, which does not rule out eventual broader availability. The timeline and conditions under which OpenAI might expand access to its new cybersecurity model remain undisclosed.
What This Means
The convergence of OpenAI and Anthropic on restricted access models signals growing industry consensus that certain AI capabilities pose sufficient security risks to warrant controlled distribution. This approach differs from each company's general model release strategy: while both companies maintain open or semi-open release policies for general-purpose models, specialized cybersecurity capabilities are receiving more cautious handling. The $10 million in credits for OpenAI's pilot suggests these capabilities remain valuable for legitimate defensive security work, but the restriction indicates both companies assess the offensive application risks as substantial enough to limit access during development and safety testing phases.
Related Articles
Chinese Models Kimi K3 and GLM-5.3 Close In on GPT-5.5 and Claude Opus 5, New Analysis Finds
A new industry analysis argues the performance gap between Chinese and Western AI models has narrowed to single-digit differences on broad benchmarks. Moonshot's Kimi K3 and Zhipu's GLM-5.3 now trail OpenAI and Anthropic's top models by only a few points on the Artificial Analysis Intelligence Index, with a clear Western edge remaining only in abstract reasoning, output reliability, and offensive cybersecurity capability.
OpenAI Launches 'Private Safety Processing' to Detect Misuse Without Storing Enterprise Data
OpenAI has built a system called Private Safety Processing that detects misuse patterns across multiple interactions without storing customer inputs or outputs. The company says it only receives narrow safety signals—type and severity of activity—while data stays encrypted on customer infrastructure.
OpenAI Previews 'Private Safety Processing' to Detect Abuse Without Retaining Customer Data
OpenAI is previewing Private Safety Processing to select customers, an automated system that monitors for misuse across multiple sessions without retaining any customer data. The move directly contrasts with Anthropic's July policy allowing 30-day data retention for 'covered models' like Fable.
Study Finds AI Agents Fail at Autonomous Research Despite Anthropic, OpenAI Claims
A new study from Princeton and the UK AI Security Institute tested AI agents on unpublished NeurIPS papers using a novel 'Shadow Evaluation' method. Both Claude Opus 4.8 and GPT-5.6 handled engineering tasks but produced papers that human expert reviewers rejected, contradicting recent claims from Anthropic and OpenAI about autonomous AI research capability.
Comments
Loading...