The company's Mythos model will be accessible to Amazon, Apple, Cisco, Google, JPMorgan Chase and Microsoft for cyber defence purposes in a bid to slow the progress of hackers in the AI arms race.

Mythos is not yet ready for public launch because of how it can be abused by cybercriminals and spies - a prospect that has caused alarm in both Washington and Silicon Valley.

Experts say that the speed and scale of AI looking for vulnerabilities greatly exceed human capabilities and could herald a revolution for the cybersecurity sector.

For example, a single AI agent could scan for vulnerabilities and take advantage of them faster and more persistently than hundreds of human hackers.

Anthropic's Logan Graham told CNN: "We did not feel comfortable releasing this generally. We think that there’s a long way to go to have the appropriate safeguards."

It has also been revealed that Anthropic has briefed senior officials "across the US government" on Mythos' full offensive and defensive cyber capabilities.

Anthropic chiefs are hopeful that the selected release of Mythos to major companies will help to level the playing field with hackers and prevent major security flaws in widely used internet browsers and operating systems before public release.

Graham told CNN: "If models are going to be this good — and probably much better than this — at all cybersecurity tasks, we need to prepare pretty fast.

"The world is very different now if these model capabilities are going to be in our lives."

Anthropic claims that Mythos has already produced promising results by finding "thousands" of previously unknown software vulnerabilities - a rate that far outpaces human experts.

These types of software flaws can be very tricky for human researchers to discover and are coveted by both spy agencies and cybercriminals for conducting stealthy hacks.