Wednesday, April 29, 2026

Google has inked a contract allowing the US Department of Defense to run its AI models on classified networks for "any lawful government purpose." The deal also lets the Pentagon request adjustments to safety filters — a concession that Anthropic refused to make, resulting in the AI safety company being flagged as a DoD "supply-chain risk." Google's contract carves out exceptions for domestic mass surveillance and autonomous weapons without "appropriate human oversight."

Alphabet is committing $10 billion immediately at a $350 billion valuation, with up to $30 billion more tied to performance milestones, plus 5 gigawatts of TPU computing capacity over five years. The announcement coincides with Anthropic reaching a $30 billion annualized revenue run-rate in early April — up from $1B at end-2024 and $9B at end-2025. Amazon has separately committed $5B as part of a broader infrastructure agreement worth up to $100B.

Anthropic has quietly released Claude Mythos to a limited set of partners, describing it as a "step change in capabilities." The model's most headline-grabbing trait: it can reportedly surpass all but the most skilled humans at finding and exploiting software vulnerabilities and has already flagged thousands of high-severity CVEs in testing. Due to dual-use concerns, broader access is blocked while Anthropic works with select security organizations to evaluate risks. A full public launch is expected in Q2 2026. Mythos is also available as a preview on Google Cloud Vertex AI.

OpenAI rolled out GPT-5.5 on April 23 to Plus, Pro, Business, and Enterprise users in ChatGPT and Codex. The model is positioned as the company's strongest yet for professional, multi-step work: coding, online research, data analysis, and document creation. GPT-5.5 Pro and GPT-5.5 Thinking variants are also available on eligible plans. An API release is coming "very soon." OpenAI's annualized revenue has now surpassed $25 billion, with reports of early IPO preparation targeting late 2026.

The four largest companies by market cap report Q1 2026 earnings after the close today — the first time all four have reported on the same day. Analysts have one central question: is AI monetization keeping pace with capex? The four hyperscalers committed a combined ~$645 billion to AI infrastructure in 2026, up 56% year-over-year. Microsoft's last quarter alone saw capex hit $29.88B (+89% YoY). Key AI watch metrics: Microsoft 365 Copilot seat growth, Google Cloud Gemini adoption, AWS growth vs. 20% benchmark, and Meta's advertising ROI from AI targeting.

Meta announced a 10% workforce reduction and Microsoft offered voluntary buyouts for the first time in its 51-year history, together totaling roughly 20,000 roles. Both companies explicitly cited AI-driven productivity gains as enabling smaller headcounts. The concurrent announcements have amplified concern that the AI-driven labor crisis in tech is no longer theoretical — and that even well-paid knowledge workers are not immune. Apple CEO Tim Cook separately announced he will step down on September 1, 2026, adding leadership uncertainty ahead of Apple's earnings Thursday.

Built on Gemini 3.1 Pro, the two new agentic research tiers — Deep Research (speed-optimized) and Deep Research Max (comprehensive, async-first) — are now in public preview via the Gemini API. Standout additions include MCP support for connecting to private data sources, and native chart and infographic generation so research outputs arrive pre-visualized. Deep Research Max is explicitly designed for background batch workflows, such as nightly due-diligence reports. Google Cloud rollout follows the API preview.

Apple has confirmed a rebuilt, AI-powered Siri is launching in 2026 — running on Google's 1.2 trillion-parameter Gemini model via Apple's Private Cloud Compute infrastructure. The arrangement preserves Apple's privacy-first positioning while outsourcing the raw model intelligence to Google. The partnership is an unusual public admission that Apple's in-house AI capabilities require a flagship partner. Details on integration depth (on-device vs. cloud inference split) remain sparse ahead of Apple's earnings call Thursday.

Cursor's redesigned Cursor 3 splits the Composer view into parallel agent lanes — run a refactor agent and a test-writing agent simultaneously and watch diffs stream side by side. The release is a direct answer to Claude Code, which topped a Pragmatic Engineer survey of 906 engineers with a 46% "most loved" rating and contributed to roughly 4% of all public GitHub commits in March 2026. OpenAI has also published an official Codex plugin that runs inside Claude Code, and Codex now claims 3 million weekly active users, up 50% in one month. The three tools are converging into a single interoperable stack.

SpaceX has reportedly landed a deal to purchase Cursor, the AI coding editor that competes with Claude Code and OpenAI's Codex. The stated rationale is accelerating model training efforts, suggesting SpaceX sees Cursor's infrastructure and developer data flywheel as strategic. The deal would be a notable exit for the Anysphere team and raises questions about whether Cursor's enterprise customers — many in regulated industries — are comfortable under SpaceX ownership. The acquisition has not been formally confirmed by either party.