Frontier Threats Red Teaming For Ai Safety

Summary

Anthropic CEO Dario Amodei also highlighted this topic in recent Senate testimony. With that context, they were pleased to advocate for and join in commitments announced at the White House on July 21 that included “internal and external security testing of AI systems” to guard against “some of the most significant sources of AI risks, such as biosecurity and cybersecurity.” However, red teaming in these specialized areas requires intensive investments of time and subject matter expertise. The team believe that improving frontier threats red teaming will have immediate benefits and contribute to long-term AI safety. Frontier threats red teaming requires investing significant effort to uncover underlying model capabilities. Following a well-defined research plan, subject matter and LLM experts will need to collectively spend substantial time (i.e.

Over the past six months, they spent more than 150 hours with top biosecurity experts red teaming and evaluating their model’s ability to output harmful biological information, such as designing and acquiring biological weapons. The experts used a bespoke, secure interface to their model without the trust and safety monitoring and enforcement tools that are active on their public deployments. The first is that current frontier models can sometimes produce sophisticated, accurate, useful, and detailed knowledge at an expert level. However, they found indications that the models are more capable as they get larger. The team also think that models gaining access to tools could advance their capabilities in biology. If unmitigated, they worry that these kinds of risks are near-term, meaning that they may be actualized in the next two to three years, rather than five or more. At the end of the project, they now have more experiments and evaluations they'd like to run than they started with.