Breaking

Anthropic Alignment Lead Gives >10% AI Extinction Estimate as U.S. Bill Odds Stay at 9%

September 9, 2026 at 11:58 EDT

Policy & Regulation
Security
Foundation Models

Anthropic Alignment Science Lead Evan Hubinger says he sees a greater-than-10% chance that AI could kill all humans within the next decade, while a Polymarket contract puts the chance of a U.S. AI safety bill becoming law before 2027 at 9%.

September 9, 2026 · AI safety

Warnings rise as U.S. bill odds stay low

An Anthropic alignment lead puts his personal estimate of catastrophic AI risk above 10%, while the market for a U.S. AI safety bill before 2027 shows 9%.

The same market, shown to scale

Yes and No are opposite outcomes in the bill market. The percentages are a market snapshot, not a scientific risk estimate.

Yes

91%

U.S. bill odds

>10%

Personal risk estimate

2027

Legislative deadline

Why it matters

Insiders are describing advanced-AI risk in unusually direct terms while labs expand public safety programs.

What remains uncertain

The >10% figure is personal, and the market tracks legislation rather than the underlying technical risk.

The policy gap

Technical warnings and statutory action are moving on different clocks.

Alignment work continues, but the legislative window closes before 2027.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.