Pentagon drops $54B on AI drones (plus vibe coding chaos)

The Pentagon is seeking $54 billion for autonomous drone warfare in 2027, a 24,000% budget increase that represents the largest commitment to autonomous military systems in history. Despite the massive investment, AI safety experts and former officials warn that both the military and AI companies are unprepared for the risks of autonomous weapons systems.

Anthropic's talk promotes 'vibe coding' - using Claude AI for production code with minimal oversight - but faces developer pushback over the non-deterministic nature of LLMs versus traditional compilers and concerns about testing rigor and learning fundamentals.

Anthropic

Anthropic fixed three distinct bugs in Claude Code that caused apparent quality degradation, including a default reasoning effort reduction, a prompt caching bug that erased thinking history, and an overly restrictive verbosity limit. The company is implementing stricter testing protocols and has reset usage limits for all subscribers.

Escape Analysis

Lovable's $6.6B vibe coding platform left critical vulnerabilities unpatched for 48 days, exposing thousands of user records and highlighting a systemic crisis: 40-62% of AI-generated code contains security flaws, but market incentives reward growth over security as AI-generated code approaches 60% of all new development.

Veracode

Veracode's analysis of 100+ AI models found that AI-generated code contains security vulnerabilities in 45% of cases across major programming languages, with larger models showing no security improvement. The findings indicate that AI-generated security risks are likely already present in production codebases.