OpenAI has launched a special version of ChatGPT, focused on giving it extra hacking capabilities.
The release is intended to allow cyber security professionals to use the system to identify potential attacks and protect against them.
GPT‑5.4‑Cyber is a model that is “purposely fine-tuned for additional cyber capabilities and with fewer capability restrictions”, OpenAI said in its announcement. Among other changes, it is less likely to refuse to find exploits that might be useful to hackers.
OpenAI’s release came very shortly after Anthropic released a new model, Claude Mythos, that functions in much the same way. That release – and the apparent power of that model – led to fears that artificial intelligence could break much of the security of the internet by finding previously undiscovered weaknesses.
The company said that it would only give access to GPT‑5.4‑Cyber to trusted organisations, who must go through a vetting process. But it said that it was intending to “make these tools as widely available as possible while preventing misuse” and that it hoped it would be available to “those responsible for protecting critical infrastructure, public services, and the digital systems people depend on every day”.
It said that it was building automated systems to verify the people who have legitimate need to have access to such systems. “This allows us to expand access based on evidence and real signals of trust, rather than relying on manual decisions,” the company wrote.
“We don’t think it’s practical or appropriate to centrally decide who gets to defend themselves. Instead, we aim to enable as many legitimate defenders as possible, with access grounded in verification, trust signals, and accountability.”
OpenAI said that rolling the launch out in this way means that cybersecurity experts will be able to check their systems are ready for the more broad deployment of new, more powerful and generally available models.
