Anthropic sends junior staffer to EU Parliament hearing, officials furious

Full text

BRUSSELS — European policymakers criticized AI company Anthropic on Tuesday for failing to send a senior company official to address Brussels’ concerns about risks posed by advanced AI capabilities.

Anthropic appointed Donny Greenberg, a technical employee who joined the company in April, to address the European Parliament via video from New York, rather than its head of public policy, Sarah Heck, as lawmakers had requested.

“I’m a technical person, not a policy person,” Greenberg stressed in his opening remarks.

The testimony was the first of its kind in Brussels since the U.S.-based company’s early April decision to restrict its cyber-capable model, Mythos, to a set of trusted American firms. EU institutions have been desperately seeking access to Mythos and Fable, both capable of finding and exploiting software vulnerabilities, thereby putting critical infrastructure at risk.

“I guess you can expect a lot of policy-related questions that perhaps you won’t be able to answer as a technical guy,” said Dutch Greens European Parliament lawmaker Kim van Sparrentak. “And I think Anthropic could have known that.”

After the hearing, van Sparrentak told POLITICO: “It’s clear from this hearing Anthropic doesn’t care about Europe.”

Greenberg joined the AI leader when it acquired his company, Runhouse, earlier this year. He has since been working on Glasswing, the company’s effort to share Mythos with trusted cyber defenders.

“A lot of [lawmakers] would have liked to speak to the political level,” said the Greens’ Anna Cavazzini, the chair of the Parliament’s internal market committee. Meanwhile, Dutch Greens European Parliament lawmaker Reinier van Lanschot told Greenberg: “I find it a bit disappointing that in this exchange we cannot touch on the policy side.”

Cavazzini later warned Anthropic that it should brace for a “next exchange of views we’re definitely having.”

Other lawmakers, meanwhile, focused on Greenberg’s virtual appearance and an impression in the room that he was reading his answers from his screen.

“Are you answering our questions, or are you just following what your AI model is telling you to say?” asked Spanish conservative lawmaker Pablo Arias Echeverría.

Greenberg fired back: “I should take that as a compliment since we’re very proud of Claude.”

In his answers, Greenberg said the rise of cyber-capable models shows AI is a dual-use tool, and that now is the time for companies to boost their cyber resilience. He also said that Anthropic is working with the Commission’s AI Office and the EU’s cybersecurity agency, ENISA, to enable that.

The hearing ended as awkwardly as it had kicked off, with Greenberg no longer visible on screen when chair Cavazzini thanked him for his appearance.

The question of if, how and by whom new advanced artificial intelligence models are tested has become a delicate geopolitical and business question as the AI race heats up.

At another meeting earlier in the day, a representative of Open AI, one of Anthropic’s chief rivals, told lawmakers it was working to avoid staged rollouts of its new models. Meanwhile on Tuesday, Google DeepMind CEO Demis Hassabis called for the U.S. to create an industry-funded standards body to test frontier AI models and set standards.