The leading open-weight AI models are closing the gap on the frontier in cyber capabilities, an analysis found.
The UK’s AI Safety Institute found the most capable open models were four to seven months behind the best closed ones.
Open-weight has advantages: Models can be modified for users’ needs and allow for easier collaboration.
But their guardrails can also be removed, making AI’s powerful cyber and biological capabilities more accessible to bad actors.
Boko Haram militants in Nigeria are using (mostly closed) AI to plan attacks and gain tactical advantages, The New York Times reported.
Open models “potentially present a higher risk,” Anthropic’s CEO wrote recently, although he suggested that all powerful models, not just open ones, should be tested for safety.
