JadidJust-in AI Developments, Incidents & Deadlines
May 2026
SMTWTFS12345678910111213141516171819202122232425262728293031

AI news, Monday, May 4, 2026

8 stories

TLDR

  • Pentagon signs classified AI deals with 7 companies (May 1) — OpenAI, Google, Microsoft, Amazon, Oracle, Nvidia, SpaceX; Anthropic labeled "supply chain risk" after refusing. 580+ Google employees push back today.
  • DALL-E 2 & DALL-E 3 retire in 8 days (May 12) — API requests will return errors; migrate to gpt-image-1.5 now or your image pipelines break.
  • Google Gemini CLI CVSSCommon Vulnerability Scoring SystemA standard 0 to 10 severity score: 0.1–3.9 low, 4.0–6.9 medium, 7.0–8.9 high, 9.0–10 critical.FIRST ↗ 10.0 RCERemote code executionA flaw that lets an attacker run malicious code on a target's computers or network.Cloudflare ↗ patched (Apr 30) — Max-severity RCE in CI/CD headless mode; fix breaks --yolo workflows.
  • EU AI Act Omnibus trilogue collapsed (Apr 28) — Next session May 13; Aug 2 high-risk deadline unchanged.
  • Big Tech AI layoffs hit 95K in 2026 (May 4 analysis) — Meta 8K effective May 20, Microsoft ~7% US voluntary buyouts; analysts frame it as AI capex funded by headcount cuts.

Action Items

Things your engineering team should look at or do this week.

  • Migrate DALL-E 3 image calls before May 12 (8 days) — After the deadline, requests to /v1/images/generations with "model": "dall-e-3" or "dall-e-2" return an error, not a fallback. Replace with gpt-image-1.5 or gpt-image-1-mini. The endpoint response schema has changes — test in staging now. (OpenAI Community | Migration guide)
  • Update Gemini CLI to ≥0.39.1 and audit CI pipelines today — The CVSS 10.0 patch removes implicit workspace trust in headless mode and drops --yolo auto-approval. GitHub Actions using run-gemini-cli or any .gemini/ directory-based config need to be reviewed — the fix may break your existing workflows. (The Register)
  • DeepSeek V4 Pro promo floor expires today (May 5) — The $0.435/$0.87 launch rate ends tonight; a 75%-off promotional rate then applies through May 31. Run benchmarks now if you want to compare against the floor rate.
  • EU AI Act high-risk compliance — Aug 2 deadline still stands — The Omnibus relief proposal failed its April 28 trilogue with next attempt May 13. If Omnibus isn't formally adopted before August 2, the original high-risk AI obligations kick in on schedule. Continue compliance prep — don't wait for May 13. (IAPP)

APIs & Pricing

  • DALL-E 2 & DALL-E 3 retire May 12 (announced Nov 2025, deadline in 8 days) — OpenAI is consolidating image generation into gpt-image-1 / gpt-image-1.5. After May 12, the /v1/images/generations endpoint will reject any call with the dall-e-2 or dall-e-3 model strings — no silent fallback. DALL-E 3 has seen an 80% drop in usage share as Imagen 3 and FLUX displaced it. Migrate now; the new endpoint response format differs from DALL-E 3. (OpenAI Deprecations | Community thread)

Security

  • Google Gemini CLI: CVSS 10.0 RCE patched (Apr 30) Agents — Researchers found that placing malicious content in a local .gemini/ directory or a project's environment variables could force Gemini CLI to execute arbitrary code before its sandbox initialized — effectively bypassing all security controls in headless/CI mode. Patched in @google/gemini-cli v0.39.1 and v0.40.0-preview.3. The fix also removes --yolo mode's blanket tool auto-approval and tightens headless workspace trust, which may break existing CI/CD pipelines that rely on that behavior. (The Register | The Hacker News | CSO Online)

Industry & Business

  • Pentagon signs classified AI deals with 7 companies; freezes out Anthropic (May 1) Policy — The DoD signed agreements with OpenAI, Google, Microsoft, Amazon, Oracle, Nvidia, SpaceX, and Reflection AI to deploy AI in classified networks for "any lawful purpose" including battlefield decision-making. Anthropic, which refused to allow Claude to be used for autonomous weapons and mass surveillance without restrictions, was labeled a "supply chain risk" — a designation previously reserved for companies tied to foreign adversaries. A federal judge has blocked the government's most aggressive enforcement steps, and Anthropic CEO Dario Amodei met White House Chief of Staff Susie Wiles last month. (CNN Business | Washington Post | Defense News)

  • 580+ Google employees (including DeepMind researchers) push back on Pentagon deal (May 4) Policy — An open letter to CEO Sundar Pichai cited risks from lethal autonomous weapons, mass surveillance, and the inability to oversee classified work. Google responded with a memo saying it "proudly" works with the U.S. military. Fortune reports today that, unlike 2018's Project Maven walkout which did force a policy reversal, employee leverage has significantly eroded — threats of resignation no longer carry the same weight inside Mountain View. (Fortune | The Next Web)

  • Big Tech layoff wave deepens: 95K tech jobs cut in 2026 (May 4 analysis) Hiring — Meta's 8,000 job cuts take effect May 20 (recruiting and HR absorb 35-40% of cuts); Microsoft offered voluntary buyouts to ~7% of US staff. Amazon cut 16,000 corporate roles in Q1 while reporting 24% AWS growth. Analysis published today asks whether $725B in collective Big Tech AI capex for 2026 is being directly funded by headcount reductions — the math is stark: human salaries are the only cost flexible enough to partially offset data center buildout in real time. Nearly 48% of Q1 tech layoffs have been explicitly attributed to AI automation. (Invezz | CNBC | Washington Post)

  • IBM CEO Study: AI to make 48% of operational decisions without humans by 2030 (May 4) — A new IBM Institute for Business Value global survey of CEOs finds they expect AI to handle nearly half of all codifiable operational decisions autonomously within four years, up from 25% today. CEOs report actively restructuring C-suite roles now to align incentives with AI-led execution, not just AI-assisted. (IBM Newsroom)

AI in Production

  • Virginia data center grid near-miss: 9 facilities offline simultaneously (May 3) Infra — A data center industry veteran described a recent near-miss in Northern Virginia where 9 data centers went offline or switched to backup power at once; when the grid recovered, it expected the load to be there, triggering an over-frequency event that nearly cascaded into rolling blackouts. The incident comes as expert warnings about gigawatt-scale AI buildouts outpacing grid capacity grow louder — 7 GW of planned US data center capacity is already delayed due to interconnection backlogs. (24/7 Wall St.)

Policy & Safety

  • EU AI Act Omnibus trilogue fails; August 2 deadline unchanged (Apr 28) — The April 28 political trilogue between the European Parliament, Council, and Commission broke down over whether AI embedded in products already governed by sectoral safety law (medical devices, industrial machinery, cars) should be exempted from the AI Act's additional requirements. Next trilogue is May 13. If no formal Omnibus adoption happens before August 2, all original high-risk AI obligations take effect as written. Organizations deploying AI in employment, education, or critical infrastructure should not pause compliance work while waiting for May 13 news. (IAPP | DLA Piper | Morrison Foerster)

Type to search · ↑↓ to move · enter to open