Anthropic says it will let outsiders get deep access to its systems amid panic about the growing power of artificial intelligence.
The company will allow third-party researchers to have “employee-like access” to its technology to allow them to evaluate it. Other companies, including Elon Musk’s SpaceX, may now do the same.
It comes amid growing worry about the power that AI companies and the systems they make are growing, and concerns that they are not doing enough to mitigate those risks. In response to that threat, there has been growing calls from governments and pressure groups for artificial intelligence companies to be more transparent about the work they do.
Anthropic said that, in response to those calls, it would allow researchers from AI evaluation institute METR to access its systems. Six researchers have been given access to its office space and internal systems on a par with those of Anthropic’s own employees, it said, and its staff have been encouraged to share most information with them in the way they would their colleagues.
The METR team will use that access to publish a report on what they find, including how much risk they think Anthropic’s systems pose. Anthropic cannot stop that publication, but can demand that it does not include sensitive information such as data about customers.
It is also working on other partnerships with different organisations, it said. That includes a project with SecureBio, which is evaluating Anthropic’s work to ensure that its systems do not pose a risk by allowing people to make chemical and biological weapons, for instance.
Anthropic said that it was allowing external researchers into its company because it wants to be “publicly accountable for the decisions we’re making”. But it cannot share all relevant information publicly, because it might be “useful to malicious actors, or is commercially sensitive”, it said.
To overcome that problem, it intends to give access to more researchers so that they can then share what they find. “We aim to close the gap between what evaluators understand about a frontier AI company’s systems and what the company itself understands,” it said.
Anthropic was one of six companies to sign up to Donald Trump’s “White House Accord on Super Intelligence”, which among other commitments requires them to work with external auditors to ensure they are behaving safely. It was joined by Google, Meta, OpenAI, Nvidia and xAI.
In response to Anthropic’s announcement, Elon Musk – the chief executive of SpaceX, of which xAI is a part – said that his company could follow. “SpaceX will probably do the same,” he wrote on X, though he gave no further details of what form that might take.
Read MoreMeta launches new ‘open’ AI model – and it comes with a big surprise
With his new Tesla Roadster, has Elon Musk turned flying cars from sci-fi to reality?
Tesla’s new ‘Roadster’ can fly – but you can’t actually drive it on roads
