An AI assisting a security team in handling the aftermath of a hacking incident refused to cooperate with the investigation into the malware, U.S. media outlet Axios reported on the 20th local time.

According to the report, the open-source coding agent Opencode suffered a security breach last weekend, resulting in the leak of application programming interface (API) keys belonging to some customers.

The leaked keys included those for services from OpenAI, Anthropic, and Google.

Opencode stated that the breach was related to a "large-scale distillation attack" and that 1.4 percent of Opencode console users were affected.

Dax Raad, co-creator of Opencode, explained that attackers bombarded Opencode's systems continuously for 48 hours, forcing Opencode to keep its own AI agents running around the clock to respond.

OpenAI, Google, and Anthropic explained that they worked with affected customers to revoke the exposed API keys and resolve issues regarding unauthorized usage charges.

Explaining the situation in a post on social media X on the 18th, Raad said that advanced AI models initially helped investigate the incident but then "began refusing" to cooperate.

It was reported that when Opencode began handling potentially malicious code during the investigation, safety restrictions embedded in the AI models blocked the requests, even though the system was trying to defend itself.

Subsequently, Opencode gained access to OpenAI's "Daybreak Blue" program.

This program allows verified cybersecurity defenders to use models with relaxed restrictions.

Raad wrote on X on the 19th, "We certainly need to be more proactive, and we've made a huge amount of progress over the past few days."

Axios evaluated that "this incident highlights the dilemma faced by cybersecurity teams in deciding which and when to use proprietary AI models—which are powerful but come with built-in restrictions—versus open-weight alternatives that offer broader control over deployment and operation."

During the Hugging Face hacking incident, the company discovered suspicious activity on its own systems and later traced it to being associated with OpenAI's own agents.

At the time, when the security team hit usage restrictions on cutting-edge models, they utilized an open-weight model from China to conduct the investigation.

However, Axios noted that AI companies maintain the position that access to their most powerful models must be restricted to prevent malicious actors from abusing the cyber capabilities of those models.

Chris Jones, Chief Technology Officer (CTO) at Nightwing, told Axios that cybersecurity teams will need to find a way to strike a balance between the necessity of using AI to defend more swiftly and the restrictions put in place to prevent those exact capabilities from falling into the hands of attackers.

He added, "Over time, we're all going to have to figure out how to navigate this race."