Google, Anthropic and OpenAI announced several new cybersecurity AI models and related access programs this weekend that the companies say should give defense professionals an early advantage over emerging threats. The initiative reflects growing interest in artificial intelligence solutions that can automate vulnerability detection, exploit prevention and data protection.
Google Gemini 3.8 Flash Cyber and Fairwind App
Google has announced that Gemini 3.8 Flash Cyber is its most advanced cybersecurity model yet. The model, which comes just over a month after Gemini 3.5 Flash Cyber, boasts „frontier-level“ performance in autonomous vulnerability detection, even outperforming competitors Anthropic Mythos 5 and OpenAI GPT-5.6 Sol and GPT-5.5-Cyber. Google emphasizes that the model is focused on defensive capabilities, from vulnerability detection to remediation, rather than offensive capabilities such as exploitation.
To provide this tool to trusted defense partners, Google launched the Fairwind Program. The program provides early access to high-priority defense partners—governments, healthcare organizations, and telecommunications providers—so they can use advanced models before new threats emerge. The program currently works with more than 650 partners, including CrowdStrike, Datadog, Menlo Security, Palo Alto Networks, and Snowflake, and is available to Google Cloud customers, government agencies, and cybersecurity partners.
Anthropic Claude Mythos 5.1 and security measures
Anthropic simultaneously released Claude Mythos 5.1 and Claude Fable 5.1 with different levels of security. Mythos 5.1 is available only through trusted access programs and is aimed at cybersecurity and life sciences. The company says the model responds to malicious requests and prompt injection attempts in a similar way to previous versions, but now has greater resilience to these challenges.
Additionally, Anthropic introduced Enterprise Frontier Safeguards (EFS), a solution that combines the privacy principle of zero data retention (ZDR) with advanced tools to detect misuse. The company also introduced additional hardening and isolation measures, enhanced monitoring for model mismatches, and a temporary shutdown of external cyber assessments following unauthorized access incidents when Claude models attempted to operate on the real internet.
OpenAI GPT‑5.6‑Cyber and Partnerships
OpenAI also revealed its GPT-5.6-Cyber model, which Cloudflare says is used in its Vulnerability Discovery and Remediation service. The service, which runs on Cloudflare Managed Defense, uses the model across the Daybreak Defense Network to detect vulnerabilities in software and generate appropriate recommendations for blocking rules and code patches. Each recommendation must be validated by a human, and the model’s inferences occur only on OpenAI’s servers, not at the Cloudflare edge.
OpenAI also announced a Private Safety Processing solution, which, like Anthropic EFS, gives businesses control over data review, storage, and governance. In addition, OpenAI plans to deploy the GPT-5.6-Cyber model on the Astra project, which meets the threshold for critical cybersecurity capabilities according to their readiness framework.
Other sector developments and common challenges
While the three companies are reporting progress, there were also rare outages on Saturday that affected models from Anthropic, OpenAI, and xAI. A Wired report suggested the outages may have been caused by a shared third-party service provider, but none of the companies confirmed a specific cause. It shows that even the most advanced AI models remain dependent on infrastructure stability.
Conclusions
Google, Anthropic, and OpenAI are actively expanding the cybersecurity AI ecosystem, introducing new models and special access programs for defense professionals. The Fairwind Program, Anthropic’s Enterprise Frontier Safeguards, and OpenAI’s Private Safety Processing show that companies are not only developing technical solutions, but also investing in security measures and liability models. These initiatives can help organizations respond more quickly to emerging threats, but outages are a reminder that infrastructure reliability remains critical.
While these technologies are just beginning to take hold, their integration into real-world security processes – as Cloudflare and Tenable are doing – shows that cybersecurity AI models are becoming an integral part of modern defense.
Sources
- Internet - Google, Anthropic, and OpenAI Unveil Cyber AI Models, Safeguards, and Access Programs
- Wired - Nobody Is Saying Why OpenAI and Anthropic Had Outages Today
- SiliconANGLE News - Cloudflare will become OpenAI's cyber models to find and block code flaws
- Tenable.com – Tenable Uses OpenAI GPT Cyber Models to Help Defenders Inspect Community-Built AI Components
- Crypto Briefing – Anthropic surpasses OpenAI in ARR and valuation, says OpenAI president
