OpenAI’s newest cybersecurity model is landing in Europe as the country pushes for access to the same AI tools that could expose weaknesses in its banks, companies, and critical infrastructure.
The company announced Thursday that it is rolling out a limited preview of GPT-5.5-Cyber, a more permissive version of its latest AI model for vetted defenders. The launch comes as European finance officials press Anthropic to grant local organizations access to its rival, the Mythos model, warning that EU companies could fall behind US peers in preparing for AI-assisted cyber threats.
According to OpenAI, GPT-5.5-Cyber is not necessarily more powerful than the standard GPT-5.5 model. Instead, it is designed to be more permissive for authorized security work such as vulnerability identification, malware analysis, reverse engineering, and patch validation.
“GPT‑5.5‑Cyber lets a smaller set of partners study advanced workflows where specialized access behavior may matter,” OpenAI said in a blog post. The company added that the model is intended to support “authorized red teaming, penetration testing, and controlled validation.”
A more open rollout than Mythos
While Anthropic limited Mythos access to roughly 40 organizations through its Project Glasswing initiative, OpenAI is taking a broader approach.
Under OpenAI’s Trusted Access for Cyber (TAC) program, verified defenders can gain access to versions of GPT-5.5 with fewer restrictions than the publicly available model. The highest tier of that program unlocks GPT-5.5-Cyber itself.
Axios reported that a source familiar with the model’s performance said GPT-5.5-Cyber is “roughly on par with Mythos.”
According to OpenAI, defenders approved through TAC can use the models for tasks including secure code review, malware analysis, vulnerability triage, and detection engineering. However, the company says safeguards still block malicious activity such as credential theft, stealth operations, or malware deployment.
AI models are getting alarmingly good at hacking tasks
The rollout lands amid growing concern in Washington and Silicon Valley over how capable frontier AI systems are becoming at cyber offense and defense.
Axios noted that recent testing by the UK AI Security Institute showed GPT-5.5 successfully completed a simulated 32-step corporate cyberattack in two out of ten attempts, while Mythos succeeded in three out of ten. Before Mythos, no AI model had reportedly passed the test.
The growing sophistication of these models has sparked discussions inside the US government. CNBC reported that senior officials, including Jerome Powell and Scott Bessent, recently met with banking executives to discuss the implications of Anthropic’s Mythos release.
Meanwhile, OpenAI says stronger identity verification and monitoring will be central to expanding access safely.
“As stronger identity and organization verification, approved-use scoping, and misuse monitoring improve, we expect access to broaden over time,” the company said.
The company also confirmed that beginning June 1, 2026, users accessing its most permissive cyber models will need phishing-resistant account protections through Advanced Account Security or approved enterprise authentication systems.
Also read: The move comes just weeks after rival Anthropic drew intense attention with its own cyber-focused model, Claude Mythos Preview.
The post OpenAI Opens Cyber Model to EU While Anthropic Keeps Mythos Restricted appeared first on eWEEK.
No Responses