AI News Briefing — May 10, 2026: LLMs Corrupt Documents at Scale, Cloudflare Cuts 1,100 Amid AI Surge, Musk v. Altman Trial Heats Up

7 Top Stories

1. Study: Even Frontier LLMs Corrupt 25% of Documents in Delegated Workflows

A new paper titled “LLMs Corrupt Your Documents When You Delegate” introduces the DELEGATE-52 benchmark — 52 professional domains requiring deep document editing — and tests 19 AI systems. The results are sobering: even leading models including Gemini 3.1 Pro, Claude 4.6 Opus, and GPT 5.4 silently corrupt roughly 25% of document content by the end of long workflows, with smaller models degrading content even more severely. Agentic tool use did not improve results, and errors compound over longer interactions and larger documents. The paper has sparked intense discussion with over 300 upvotes on Hacker News and 120+ comments.

2. Cloudflare Lays Off 1,100 Workers as AI Usage Surges 600%

Cloudflare announced it is laying off 1,100 employees despite — or because of — a 600% increase in AI usage across its platform. The company framed the decision as strategic repositioning rather than cost-cutting, stating it is “defining how a world-class, high-growth company operates and creates value in the agentic AI era.” The move highlights a growing tension in tech: companies are scaling AI infrastructure rapidly while simultaneously reducing headcount, betting that autonomous AI agents will replace significant portions of human labor.

3. Meta’s AI Push Reportedly Making Employees “Miserable”

The New York Times reports that Meta employees are growing increasingly frustrated as the company pushes an aggressive AI transformation. Meta has begun tracking employees’ computer activity to train its AI models, plans to cut 10% of staff later this month, and is reportedly pushing teams to create so many AI agents that “others had to introduce agents to find agents, and agents to rate agents.” Some employees say they no longer see Meta as a long-term career destination, with some actively trying to get laid off to collect severance.

4. Musk v. Altman Trial Enters Week Two with Explosive Testimony

The trial between Elon Musk and OpenAI/Sam Altman continued with deposition testimony from former OpenAI board member Helen Toner, who revealed the board discussed merging OpenAI with Anthropic during “the Blip” — the brief November 2024 period when Altman was fired. Toner also confirmed the board did not review Altman’s or Brockman’s HR files before the firing, nor did they consult Microsoft or other major investors. Tasha McCauley testified about a “culture of lying and deceit” at OpenAI, while safety researcher Rosie Campbell discussed the disbanding of OpenAI’s AGI readiness team in 2024.

5. OpenAI Launches Codex Chrome Extension for In-Browser Automation

OpenAI released a Chrome extension for its Codex AI coding agent, enabling it to interact with websites and web apps where users are already logged in. The extension works through “task-specific” tab groups, allowing users to keep their active browsing separate from Codex’s automated sessions. The move signals OpenAI’s push to make Codex a general-purpose automation tool that extends beyond code generation into everyday web workflows.

6. Google Personalizes Gmail’s “Help Me Write” with Your Tone and Style

Google announced a significant upgrade to its “Help me write” AI feature in Gmail, now capable of generating emails that match the user’s personal writing tone and style. The updated tool can also pull in relevant context from Google Drive and Gmail based on the user’s prompt, making it a more contextual and personalized email drafting assistant. The rollout is part of Google’s broader effort to make its AI tools feel less generic and more tailored to individual users.

7. Sony and TSMC Form Joint Venture for Next-Gen AI Image Sensors

Sony and TSMC announced a joint venture — majority-owned by Sony — combining Sony’s sensor designs with TSMC’s advanced manufacturing to develop next-generation image sensors. The partnership explicitly targets physical AI applications in robotics and automotive industries, signaling growing investment in the hardware layer needed for embodied AI systems. The collaboration bridges the gap between sensor design and cutting-edge semiconductor fabrication at a time when AI-powered robotics is accelerating rapidly.

Trend Watch

What to Watch

• Musk v. Altman closing arguments — Expected around May 16-17, the trial’s conclusion could reshape public perception of OpenAI’s governance and Sam Altman’s leadership.
• Meta’s 10% workforce reduction — Scheduled for later this month, the layoffs will test whether Meta’s all-in AI bet can sustain operations with significantly fewer engineers.
• DELEGATE-52 follow-up research — With 305+ upvotes and major community interest, expect responses from model providers and follow-up studies on delegation safety within days.