Just ten days ago, Anthropic had to temporarily shut down its AI model Claude Mythos 5 and the slimmed-down Fable 5 due to security concerns from the US government. Authorities and companies remain tight-lipped about the reasons.
Myth 5 is said to have cracked the NSA
According to a new report, Mythos 5 is said to have managed to crack the security systems of the US secret service NSA within a few hours. But Anthropic denies this, as Golem writes. Meanwhile, rival OpenAI is causing a stir with an update to its GPT-5.5-Cyber model, which specializes in AI security.
Editorial recommendations
${content}
${custom_anzeige-badge}
${custom_tr-badge}
${section}
${title}
Because this new version is intended to surpass Mythos 5 when it comes to finding security holes in large code bases. In the OpenAI benchmark Cybergym, the updated GPT 5.5 Cyber not only leaves behind its predecessor (81.9 percent) with 85.6 percent, but also Mythos 5, which achieved 83.8 percent.
GPT 5.5 Cyber beats Myth 5
The Exploitgym test shows even greater improvement. While GPT-5.5 achieved a value of 25.95 percent, the new GPT-5.5-Cyber has a value of 39.5 percent, as Computerbase writes.
GPT-5.5-Cyber is part of OpenAI’s security initiative called Daybreak and should only be made available to registered partner companies. These include 30 US companies such as Cloudflare, Sophos and Palantir. In addition, there will also be partnerships with companies in Germany, Great Britain, the EU and other Western countries.
Model update also for customer projects
The special thing about the updated model is that partner companies are now no longer only allowed to use it internally, but also for customer projects. However, the AI model itself may not be made available to customers.
Apple Intelligence: In these 6 cases the AI was wrong
Incidentally, OpenAI does not assume that its AI model will face a similar fate to that of Anthropic. The company is in “continuous communication with the US government about our cyber initiative,” said OpenAI. This includes the release of GPT-5.5-Cyber as well as other models, as stated at heise.de.
Review service for open source projects
It is also interesting that OpenAI is launching a review service for open source projects together with the security consultancy Trail of Bits. The Patch the Planet project uses GPT-5.5-Cyber to check the code, fix any security holes found and test the result. Over 30 open source projects such as cURL, Go, Python and Sigstore are already on board.
Top Article
${content}
${custom_anzeige-badge}
${custom_tr-badge}
${section}
${title}
