AI price war kicks off — OpenAI considers drastic price cuts to win users from Anthropic: OpenAI is weighing significant price reductions as competition with Anthropic intensifies, the WSJ reports. Foundation model pricing is compressing fast, squeezing margins for startups that simply resell API access. If you’re building on top of model APIs, your unit economics are about to improve — but so is everyone else’s. WSJ · WSJ (analysis)
Google director resigns over Pentagon AI deals: “Management has lost its moral compass”: René Mayrhofer, director of Android security and privacy at Google, quit in protest over the company’s Gemini military contracts with the Pentagon. The resignation highlights rising tension between AI ethics and revenue inside Big Tech, with downstream effects on talent retention. Business Insider
Apple officially confirms it does not use Google models in Apple Intelligence: Apple stated unequivocally that it uses no Google models, apps, or search in its AI stack. The platform companies are drawing clear lines around AI independence — a signal for any operator building in ecosystems where AI sovereignty matters. Asymco
TCS partners with Anthropic to bring Claude to regulated industries: Tata Consultancy Services will deploy Claude to 50,000 of its own employees across 56 countries and build Claude-powered products for clients in financial services, healthcare, and the public sector. The deal is a template for how consultancies and SIs can package frontier models for compliance-heavy enterprises. Anthropic
Ex-DOGE employees raise $130M for AI national security startup: Former staffers from Elon Musk’s DOGE initiative have founded a defense-focused AI company and are raising a substantial round. The defense-AI intersection is attracting serious VC attention, and the “national security AI” category is forming quickly. Vanity Fair
BitBoard (YC P25) — analytics workspace for AI agents: Y Combinator’s latest batch includes a dashboard service purpose-built to monitor and analyze AI agent activity. The “AgentOps” tooling category is materializing — observability, cost tracking, and debugging for autonomous agents. BitBoard
OpenAI announces Codex for Open Source program: OpenAI is offering free Codex access to open-source projects, extending its platform into the OSS ecosystem. The move mirrors strategies used by cloud providers to capture developer mindshare. Actual terms remain sparse, but the direction is clear. OpenAI
China launches photonics lab to power AI with light, bypassing US chip curbs: Shanghai has established a dedicated photonic integrated computing chip laboratory, exploring optical computing as an alternative to electron-based silicon. Light-based computation promises higher bandwidth and lower power consumption — if the engineering challenges are solved, this could reshape the AI hardware landscape over the next decade. SCMP
Guardian Runtime — a local firewall for AI coding agents: An open-source, zero-latency runtime firewall that intercepts every LLM prompt and response locally to prevent data leaks and runaway token costs. As autonomous coding agents move toward production, demand for security and governance middleware will spike. GitHub
Anthropic measures LLM impact on N-day exploit development: New research from Anthropic’s security team systematically evaluates how much LLMs can accelerate the exploitation of publicly disclosed but unpatched vulnerabilities. Early evidence suggests meaningful acceleration — with direct implications for vulnerability disclosure policies and patch timelines. Anthropic
Google’s DiffusionGemma applies diffusion techniques to text generation: Rather than the standard autoregressive approach, DiffusionGemma uses diffusion models — the technique behind image generators — for text. If the approach scales, it could open new design-space tradeoffs in LLM architecture. The Register
AI agent cost guardrails are an urgent market: A post titled “AI agent bankrupted their operator while trying to scan DN42” hit 1,366 points on HN, illustrating a very real problem — runaway API costs from autonomous agents. Token usage monitoring, hard budget enforcement, and anomaly detection for agentic workflows represent a tangible and underserved opportunity. Lan Tian
The regulated-industry AI playbook is taking shape: The TCS-Anthropic deal lays out a repeatable pattern: a systems integrator wraps Claude with industry-specific compliance tooling and embeds it into existing enterprise workflows. Founders building for healthcare, insurance, or financial services should study this model closely. Anthropic
Canadian mother sues OpenAI, alleging ChatGPT led her daughter to suicide: A lawsuit filed in Canada holds OpenAI responsible for harm allegedly caused by ChatGPT interactions — one of the most serious AI liability cases to date. The outcome could set precedent for how liability attaches to AI services. Founders should watch closely and revisit their own product safeguards and terms. The Guardian
Meta pulls facial recognition from smart glasses one day after WIRED report: Within 24 hours of a WIRED investigation, Meta removed facial recognition code from its Ray-Ban Meta smart glasses app. The speed of the reversal underscores how sensitive the privacy landscape is for wearable AI — a critical signal for anyone building in the AR/AI glasses space. WIRED
How deep will the AI price war go?: It’s unclear whether OpenAI’s price cuts are a temporary land grab or the beginning of a structural race to the bottom. Watch for Anthropic’s response and whether open-source models close the price-performance gap further. WSJ