Every AI lab thinks it's the responsible one, and safety researcher Ryan Greenblatt says that's what keeps the arms race going
Every AI lab says it's being responsible. Ryan Greenblatt says that's exactly why the race won't stop.
Why it matters
A safety researcher at Redwood Research argues that competitive dynamics between Anthropic and OpenAI—each convinced of their own caution—paradoxically accelerate risk-taking. The gap between stated safety commitment and market pressure is structural, not solvable by individual labs; international agreement is being counted on as the brake.
The key facts
10 to knowRyan Greenblatt, Redwood Research chief scientist, estimates 50-60% risk of AI takeover if current development trajectory continues
Greenblatt cites arms-race dynamics between Anthropic and OpenAI as a driver of accelerated risk-taking despite individual lab safety claims
Sam Harris comparison: Manhattan Project scientists would have halted at 10% risk; current AI pace incompatible with historical caution thresholds
Greenblatt is banking on international agreement as the mechanism to break competitive dynamics
Source: The Decoder interview reporting; no independent corroboration of risk estimate methodology provided
Ryan Greenblatt, chief scientist at Redwood Research, estimates 50–60% AI takeover risk on current trajectory
Sam Harris counterpoint: Manhattan Project scientists would have stopped at 10% risk, suggesting industry tolerance is dangerously high
Greenblatt blames race dynamics at Anthropic and OpenAI, not technical capability alone
Greenblatt counting on international agreement as mitigation
Core claim: every AI lab believes it is the responsible actor, which perpetuates the arms race
Go to the source
The Decoderthe-decoder.com
Publisher excerpt: Ryan Greenblatt, chief scientist at Redwood Research, puts the risk of an AI takeover at 50 to 60 percent if development stays on its current path. Sam Harris says that doesn't square with how fast the industry is moving, since Manhattan Project scientists would have called things off at 10…