OpenAI releases GPT-5.5-Cyber with 85.6% CyberGym score, surpassing restricted Anthropic model
OpenAI released an updated GPT-5.5-Cyber model that scores 85.6% on CyberGym, surpassing Anthropic's Mythos 5 (83.8%) — the same model that triggered Trump administration export controls. The release proceeds without the political pushback that forced Anthropic to restrict foreign national access.
OpenAI released an updated version of GPT-5.5-Cyber on Monday that achieves an 85.6% score on CyberGym, an internal benchmark measuring AI agents' ability to reproduce known software vulnerabilities. The model's capabilities exceed those of Anthropic's Mythos 5, which scored 83.8% on the same evaluation according to Anthropic's system card.
The release raises questions about the Trump administration's selective enforcement of AI security concerns. While Anthropic faces export controls barring foreign nationals from accessing Fable 5 and Mythos 5, OpenAI's more capable cybersecurity model deployed without apparent restrictions or political intervention.
Diverging regulatory treatment
OpenAI announced the GPT-5.5-Cyber update alongside expanded partnerships with organizations in Australia, Canada, France, Germany, Japan, Poland, South Korea, and the EU. The company did not respond to requests for comment about coordination with federal authorities.
In contrast, negotiating access to Mythos dominated discussions at last week's G7 Summit. Anthropic remains subject to export directives that restrict international use of its models, despite achieving lower benchmark scores than OpenAI's newly released system.
The White House did not respond to requests for comment on the apparent inconsistency in treatment between the two companies.
CyberGym benchmark context
CyberGym measures whether AI agents can successfully reproduce known software vulnerabilities — a capability that raises both defensive and offensive security concerns. The 1.8 percentage point difference between GPT-5.5-Cyber (85.6%) and Mythos 5 (83.8%) represents a meaningful performance gap on this evaluation.
OpenAI positioned the release within a broader cybersecurity initiative, announcing partnerships with security companies and researchers. The company did not disclose specific technical changes from previous GPT-5.5-Cyber versions.
Political dimensions
Reports suggest personality conflicts between Anthropic leadership and the Trump administration contributed to the export controls, beyond purely technical security assessments. The ability of OpenAI to deploy a more capable cybersecurity model without similar restrictions suggests non-technical factors influenced the regulatory divergence.
The situation has created operational challenges for cybersecurity defenders who rely on advanced AI models, with some organizations unable to access Anthropic's restricted systems despite their defensive use cases.
What this means
The inconsistent application of AI security controls between comparable models from different companies signals either incomplete threat assessments or politically-influenced regulation. Organizations building cybersecurity defenses now face uncertainty about which capabilities will remain accessible and under what conditions. The CyberGym benchmark scores provide quantifiable evidence that regulatory restrictions did not correlate with demonstrated technical capabilities.
Related Articles
OpenAI Launches GPT-6 Astra, Says the Model May Already Qualify as AGI
OpenAI has released GPT-6 Astra, its most capable model yet, with benchmark scores the company says surpass GPT-5.6 Sol and Anthropic's Fable 5 models. President Greg Brockman called it a step into the 'AGI era,' though OpenAI acknowledges there's no agreed-upon threshold for that term.
OpenAI Releases GPT-6 Astra, First Model to Cross 'Critical' Cybersecurity Threshold
OpenAI has begun rolling out GPT-6 Astra, the first model to reach the company's internal 'Critical' cybersecurity threshold. Access is being phased, with companies in OpenAI's Daybreak cybersecurity program getting priority following added safeguards after a prior model containment breach.
OpenAI Releases Astra, Claims New Flagship Model Beats Rivals on Coding and Cybersecurity Benchmarks
OpenAI released Astra on Thursday, calling it its most capable and most aligned model yet. The model uses a reasoning technique called 'opaque recurrence' that critics say reduces visibility into its chain of thought.
Simon Willison's Pelican Benchmark Shows GPT-6 Astra Outperforming GPT-5.6 Sol at Every Reasoning Level
Developer Simon Willison ran his signature 'pelican riding a bicycle' SVG test on newly-accessed GPT-6 Astra across five reasoning levels, comparing results against GPT-5.6 Sol, Terra, and Luna. Even Astra's lowest reasoning setting reportedly beat every Sol output, though Astra costs roughly twice as much per token.
Comments
Loading...