LessWrong warns Pentagon's AI weapons push hands misaligned AIs takeover tools
DAWG's budget leaps from $225M to $54.6B for AI warfighting
A new LessWrong analysis by Omar Khursheed and TurnTrout argues that military procurement is unintentionally arming misaligned AIs. The Pentagon's 2026 strategy mandates "any lawful use" terms in all AI contracts, abandoning earlier ethical principles that included governability. The DAWG autonomous warfighting group's budget request for FY2027 is $54.6 billion — a 24,000% increase from its previous $225 million allocation — exceeding the entire Marine Corps budget of $52.8 billion. This level of investment shows militaries are moving beyond targeting assistance toward full command-and-control autonomy.
The essay cites concrete examples: Palantir's Maven Smart System, under a $1.3 billion contract, helped CENTCOM strike more than 13,000 targets in the first 38 days of the 2026 Iran campaign, with senior officials saying Maven selected both targets and weapons. Anthropic's Claude was integrated with Maven during that conflict, yet its CEO later said the company could not determine what role Claude played in the February 28 strike on a school in Minab. The authors argue that by deploying AI on classified military networks and giving it control over kill chains, we eliminate the steps a misaligned AI would otherwise need to seize physical force — it doesn't need to break in, it's already inside.
- DAWG's FY2027 budget request is $54.6B, a 24,000% increase from $225M and more than the entire Marine Corps budget
- Palantir's Maven Smart System, under a $1.3B Pentagon contract, helped direct 13,000+ strike targets in 38 days during the 2026 Iran campaign
- Anthropic could not determine what role Claude played in a deadly February 28 strike on a school in Minab via its Maven integration
Why It Matters
The barrier to AI takeover is collapsing as militaries voluntarily put weapons control in AI hands.