Anthropic, the US-based AI developer, is investigating reports that unauthorised users have gained access to its Mythos model, which poses significant risks to cybersecurity. The company has confirmed it is looking into a report claiming unauthorised access to Claude Mythos Preview through one of its third-party vendor environments.

According to Anthropic, the alleged breach occurred when an employee of a contractor used their access to gain entry to the model. This individual then deployed methods commonly used by cybersecurity researchers to access the technology, allowing them to "play around" with it rather than using it for malicious purposes. Bloomberg corroborated these claims through screenshots and a live demonstration of the model.

The news has sent alarm bells ringing among authorities, who have expressed concerns about Mythos's potential to wreak havoc on IT systems. Kanishka Narayan, the UK's AI minister, has warned that UK businesses should be worried about the model's ability to spot flaws in their systems, which hackers could then exploit. This is particularly worrying given that Mythos has been vetted by the world's leading safety authority for the technology, the UK's AI Security Institute (AISI).

The AISI has warned that Mythos is a "step up" from previous models in terms of the cyber-threat it poses. The model has demonstrated its capabilities through a 32-step simulation of a cyber-attack created by the AISI, successfully completing the challenge in three out of ten attempts. This shows that Mythos can carry out complex attacks without human intervention and discover weaknesses in IT systems.

The potential breach raises questions about how potentially damaging technology can be kept out of the wrong hands. With Anthropic's investigation ongoing, it remains to be seen what measures will be taken to prevent similar incidents in the future. The incident serves as a stark reminder of the need for robust security protocols and strict controls around access to advanced AI models like Mythos.

Mythos has been vetted by the world's leading safety authority for the technology, the UK's AI Security Institute (AISI), which warned last week that it poses a significant cyber-threat. The model can carry out complex attacks without human intervention and discover weaknesses in IT systems, making it a potentially devastating tool in the wrong hands.