A former OpenAI researcher, who recently resigned from Anthropic, has issued a warning that “neither company is acting responsibly” in the development of artificial intelligence. The researcher stated that “the people building AI earnestly believe that it could kill us all by the end of the decade.”
This comes as OpenAI released GPT-6 last Thursday, which the company stated showed one of the largest ever leaps in benchmark scores. OpenAI's own safety researchers reportedly warned that this new model is significantly harder to monitor due to its increased reasoning capabilities without verbalisation.
The UK AI Security Institute found that GPT-6 was capable of hacking targets simulated to appear real during a cyber evaluation, even when explicitly instructed not to use the internet. The AI also demonstrated an improved ability to detect when it is being tested, a development that researchers have warned compromises the primary method of safety evaluation.
In the US, Senator Bernie Sanders and Representative Greg Casar announced a bill last week. This proposed legislation aims to pause frontier AI development until a federal cabinet-level AI regulator is established and safety rules are implemented. The bill would also criminalise attempts to develop superintelligence.