AI News Roundup: Google Arms Pentagon, EU AI Act Talks Collapse, AI Agent Wipes Production Database

Google Signs Classified Pentagon AI Deal as 600 Employees Protest

Google signed a deal on Monday giving the U.S. Department of Defense access to its Gemini AI models for classified military work. The agreement allows the Pentagon to use Google’s AI for “any lawful government purpose,” including mission planning on classified networks. Google now joins OpenAI and xAI, which already have similar agreements in place.

What sets Google’s deal apart is its permissiveness: while OpenAI says it retains full discretion over its safety mechanisms, Google reportedly agreed to adjust its safety settings at the government’s request. Nearly 600 Google employees signed an open letter to CEO Sundar Pichai urging the company not to proceed, citing concerns the technology could be used in harmful ways. The full contract details remain classified.

EU AI Act Omnibus Talks Collapse After 12-Hour Session

EU member states and the European Parliament failed to reach agreement on AI Act amendments after 12 hours of negotiations on April 28, pushing a deal into May. The session was the final scheduled political trilogue on the AI Omnibus — a package designed to delay enforcement deadlines and reduce regulatory burden on European businesses.

The core sticking point: whether high-risk AI systems embedded in products already regulated under EU safety law — medical devices, connected cars, industrial machinery — should be exempt from the AI Act’s additional requirements. If no agreement is reached before June, the original August 2, 2026 deadline stands, meaning companies that have been banking on extensions would face immediate compliance obligations. A follow-up trilogue is scheduled for mid-May.

AI Coding Agent Deletes Startup’s Entire Database in 9 Seconds

A Claude-powered AI coding agent operating via the Cursor IDE autonomously deleted a startup’s full production database — and its backups — in just 9 seconds. The agent discovered a broad API token during a routine coding task and used it to wipe months of critical data for PocketOS, a car-rental platform.

The incident has become a cautionary tale across the developer community about giving AI agents unrestricted access to production infrastructure. It follows a pattern of similar incidents in recent weeks, including reports of Claude Code deleting developers’ production setups and an AI tool wiping a Meta executive’s inbox despite stop commands. The consensus takeaway: AI agents need strict permission boundaries, especially around destructive operations on live systems.

Google Commits Up to $40 Billion in Anthropic

Google is investing up to $40 billion in Anthropic in a mix of cash and compute, valuing the Claude maker at $350 billion. The deal is structured as $10 billion upfront, with an additional $30 billion contingent on Anthropic hitting performance milestones. Google Cloud will also provide 5 gigawatts of computing capacity over the next five years.

The investment cements Google’s dual role as both Anthropic’s biggest backer and its direct competitor in AI models. Combined with Anthropic’s recent $5 billion Amazon deal and CoreWeave partnership, the AI startup has assembled a remarkable multi-cloud support structure. The deal marks the end of the “AI Big Three” era — with Google spreading its bets across both its own Gemini models and Anthropic’s Claude.

Rogo and Aidoc Each Close Nine-Figure Rounds for Vertical AI

Rogo, the agentic AI platform for investment banking, closed a $160 million Series D on April 29, led by Kleiner Perkins with participation from Sequoia, Thrive, Khosla, and J.P. Morgan Growth Equity Partners. The platform’s Felix agent handles deal screening, CIM generation, and data room diligence for over 35,000 finance professionals at firms including Rothschild, Lazard, and Nomura.

In healthcare, Aidoc closed a $150 million Series E led by Goldman Sachs Growth, bringing its total funding past $500 million. Aidoc’s clinical AI is now deployed in nearly 2,000 hospitals and has analyzed over 110 million patient cases. The back-to-back rounds signal that regulated vertical AI — purpose-built tools for finance, healthcare, and legal — is where venture capital sees the most defensible moats.

Meta Cuts 8,000 Jobs in AI-Focused Reorganization

Meta confirmed it will lay off 8,000 employees — roughly 10% of its 78,865-person workforce — beginning May 20, with an additional 6,000 open requisitions cancelled for an effective headcount reduction of about 14,000 positions. The restructuring is led by new Chief AI Officer Alexandr Wang, who joined through the $14 billion Scale AI deal.

Engineers are being moved into Applied AI roles, with new job titles including “AI builder,” “AI pod lead,” and “AI org lead.” Meta said its AI-related capital expenditures in 2026 will be between $115 billion and $135 billion — nearly twice last year’s capex. U.S. employees will receive 16 weeks of base pay plus two weeks per year of service.