Recent events
fr33OpenAI's GPT-5.5-Cyber outperforms Anthropic's Mythos 5 amid export control disparity
OpenAI previously released GPT-5.5-Cyber to vetted cyber defenders for critical infrastructure security, with capabilities nearly on par with Anthropic's Mythos Preview. Now, new reporting shows GPT-5.5-Cyber scored 85.6% on the CyberGym benchmark, surpassing Mythos 5's 83.8%. OpenAI released the model without facing the same export controls that restrict Anthropic's models, and the White House has not commented on the discrepancy. Reports suggest personality clashes between Anthropic and the Trump administration influenced the export directive. OpenAI expanded partnerships with organizations in Australia, Canada, France, Germany, Japan, Poland, South Korea, and the EU. The model lacks cyber safeguards and is only accessible to defenders in the highest level of its Trusted Access for Cyber Program.
Show summaryHide
OpenAI's GPT-5.5-Cyber outperforms Anthropic's Mythos 5 amid export control disparity
OpenAI previously released GPT-5.5-Cyber to vetted cyber defenders for critical infrastructure security, with capabilities nearly on par with Anthropic's Mythos Preview. Now, new reporting shows GPT-5.5-Cyber scored 85.6% on the CyberGym benchmark, surpassing Mythos 5's 83.8%. OpenAI released the model without facing the same export controls that restrict Anthropic's models, and the White House has not commented on the discrepancy. Reports suggest personality clashes between Anthropic and the Trump administration influenced the export directive. OpenAI expanded partnerships with organizations in Australia, Canada, France, Germany, Japan, Poland, South Korea, and the EU. The model lacks cyber safeguards and is only accessible to defenders in the highest level of its Trusted Access for Cyber Program.
OpenAI previously released GPT-5.5-Cyber to vetted cyber defenders for critical infrastructure security, with capabilities nearly on par with Anthropic's Mythos Preview. Now, new reporting shows GPT-5.5-Cyber scored 85.6% on the CyberGym benchmark, surpassing Mythos 5's 83.8%. OpenAI released the model without facing the same export controls that restrict Anthropic's models, and the White House has not commented on the discrepancy. Reports suggest personality clashes between Anthropic and the Trump administration influenced the export directive. OpenAI expanded partnerships with organizations in Australia, Canada, France, Germany, Japan, Poland, South Korea, and the EU. The model lacks cyber safeguards and is only accessible to defenders in the highest level of its Trusted Access for Cyber Program.