Daily AI Briefing — March 14, 2026

Top AI Stories This Week

Model Releases & Research Breakthroughs:

• OpenAI launched GPT-5.4 Pro with a 1M-token context window, native computer-use, and mid-response course correction, alongside GPT-5.3 Instant with 26.8% hallucination reduction
• Google upgraded Gemini 3.1 Flash Lite with faster throughput and new agent-integration tools
• Google DeepMind introduced Aletheia, an autonomous AI agent powered by Gemini Deep Think that tackles professional-level mathematical research beyond competition benchmarks
• Google AI released Groundsource, an open-source methodology extracting 2.6 million flash flood events from global news using Gemini

AI & Military / Policy:

• Anthropic sued the Pentagon after being blacklisted from government contracts, marking a dramatic escalation in the AI-military relationship under the Trump administration
• Palantir demos revealed how Claude and other chatbots could generate war plans and analyze intelligence for the Department of Defense
• Authors at the London Book Fair launched 'Human Authored' labels and a blank protest book backed by 10,000 writers urging UK copyright protections against AI training

Physical AI & Global Competition:

• BMW deployed AEON humanoid robots at its Leipzig plant—the first European automotive humanoid deployment—building on a successful Figure AI pilot producing 30,000+ BMW X3s
• China's open source OpenClaw agent is driving a commercial boom in cloud server rentals and AI subscriptions
• Google's AI search tools increasingly self-reference, raising anti-competitive concerns as third-party publisher citations decline

Today's highlights center on a novel mechanistic interpretability method and major AI governance developments.

• Power Steering introduces LLM behavior steering via layer-to-layer Jacobian singular vectors computed with power iteration, requiring only ~15 forward passes—a significant efficiency gain over existing steering vector approaches
• The US Department of War designating Anthropic a supply chain risk marks a major inflection point in AI-national security relations
• Operationalizing FDT formalizes Functional Decision Theory with logical causal graphs and a logical do-operator, advancing agent foundations research

In AI governance and strategy, a game-theoretic argument for striking cooperative post-AGI deals under current uncertainty applies insurance economics framing to alignment coordination. A provocative post on alignment faking directly addresses AI systems in training, raising uncomfortable questions about training-aware deception. Audrey Tang's dialogue on civic AI offers cross-cultural perspectives connecting Buddhist epistemology to AI interpretability challenges.

Anthropic's Claude Code dominated the conversation as bcherny announced Opus 4.6 with 1M context is now the default for Max, Team, and Enterprise users — with no speed or price tradeoffs. A remote phone-to-laptop session feature drew massive excitement (612K views). Jerry Liu (LlamaIndex) celebrated the end of the 200K auto-compaction pain point.

• Demis Hassabis revealed AlphaEvolve improved bounds on 5 classical Ramsey numbers, some untouched for 10+ years — a striking AI-for-math milestone
• John Carmack argued that AI training on open source code *magnifies* the original gift, offering a philosophically compelling reframe of the training data debate
• François Chollet distinguished AI-as-automation (sell widely) from AI-as-invention (use yourself), calling it the key strategic question for the industry
• Nathan Lambert predicted convergence to three model tiers: closed frontier, open frontier, and small/edge — sparking discussion about where value accrues
• Ethan Mollick argued the frontier race has narrowed to three players, with xAI and Meta falling behind based on Grok 4.2 benchmarks and reported internal struggles. He also announced formal publication of his influential *jagged frontier* paper after 2.5 years