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

AI news, Tuesday, May 12, 2026

8 stories

TLDR

  • OpenAI launches Daybreak today (May 12) — AI-native cybersecurity platform built on GPT-5.5 and Codex to rival Anthropic's Mythos; shipping with 8 major security vendor integrations.
  • Google disrupts first documented AI-built zero-dayAn attack using a flaw that was unknown before it was exploited.NIST ↗ (May 11) — Criminal group used an AI model (OpenClaw-type) to find and exploit a zero-day in a widely used sysadmin tool; Google blocked the planned mass exploitation event.
  • DALL-E 2 & 3 endpoints die today (May 12) — All API calls to dall-e-2 and dall-e-3 now fail; migrate to gpt-image-1 / gpt-image-1-mini immediately.
  • xAI retires 8 Grok models in 3 days (May 15) — grok-code-fast-1, grok-3, grok-4-fast-*, and others go dark; replacement is grok-4.3.
  • GM swaps 600 IT workers for AI-skilled hires (May 11) — Joins Coinbase, Freshworks in explicitly restructuring around AI capability gaps, not just cutting headcount.

Action Items

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

  • DALL-E 2 & 3 are dead as of today — Any pipeline still calling dall-e-2 or dall-e-3 is broken right now. Migrate to gpt-image-1 (standard) or gpt-image-1-mini (cheaper). Note: the request/response shape, resolution options, and billing model are not drop-in compatible with DALL-E 3. (OpenAI deprecations | Migration guide)
  • Migrate xAI Grok models before May 15 (3 days) — Eight models retire including grok-code-fast-1, grok-3, and the grok-4-fast-* reasoning/non-reasoning variants. Migrate code workloads to grok-4.3; image generation to grok-imagine-image. (xAI docs)
  • Read Google's GTIG AI threat report — this is the new attack surface — A criminal group used an AI model to independently discover a zero-day bypass of 2FA in a popular open-source sysadmin tool, then planned mass exploitation. First confirmed case of AI-generated zero-day in a criminal operation. If you use AI-assisted code review or threat modeling, your threat model just changed. (Google Cloud Blog | The Register)
  • Evaluate OpenAI Daybreak for your security toolchain — Launched today with GPT-5.5 and Codex agenticDescribes AI systems where a model uses tools and decides its own steps instead of following a fixed script.Anthropic ↗ capabilities for secure code review, patch validation, and dependency risk scanning. Three tiers: standard GPT-5.5, Trusted Access for Cyber (for authorized defensive work), and GPT-5.5-Cyber (for red team/pen testing). Partners include Cloudflare, CrowdStrike, Cisco, Palo Alto Networks. (The Hacker News | Business Standard)

Models & Tools

  • OpenAI Daybreak (May 12) [Security] [Agents] — OpenAI's new AI-native cybersecurity platform, launched today, positions directly against Anthropic's Mythos and Project Glasswing. Built on GPT-5.5 plus Codex agentic capabilities; three model tiers let teams customize permissiveness for defensive vs. offensive security work. Partners include Akamai, Cisco, Cloudflare, CrowdStrike, Fortinet, Oracle, Palo Alto Networks, and Zscaler. Generates and tests patches within repos and produces audit-ready evidence. (The Hacker News | Dataconomy)

APIs & Pricing

  • DALL-E 2 & 3: officially retired today (May 12) [Trending] — (Update from yesterday's action item.) As of today, the OpenAI API refuses calls to dall-e-2 and dall-e-3. The replacement is gpt-image-1 / gpt-image-1-mini; billing, resolution handling, and response shape all differ — this is not a drop-in swap. If you're still seeing errors in image pipelines, this is the cause. (OpenAI deprecations)

  • xAI: 8 Grok models retire May 15 — Three days out. Affected: grok-4-1-fast-reasoning, grok-4-1-fast-non-reasoning, grok-4-fast-reasoning, grok-4-fast-non-reasoning, grok-4-0709, grok-code-fast-1, grok-3, grok-imagine-image-pro. Recommended migration path: grok-4.3 for code and reasoning, grok-4.20-non-reasoning for fast non-reasoning, grok-imagine-image for image gen. (xAI docs)

Security

  • First confirmed AI-built zero-day used in planned criminal mass exploitation (May 11) [Trending] — Google's Threat Intelligence Group (GTIG) says it has "high confidence" that a criminal threat actor used an AI model to find and exploit an unknown zero-day in a widely used open-source system administration tool — specifically, a bypass for two-factor authentication. The actor planned to use the exploit in a mass exploitation event; Google's proactive counter-discovery appears to have prevented it. GTIG chief analyst John Hultquist called it "a moment cybersecurity experts have warned about for years." OpenClaw, an open-source AI assistant that can interface with 50+ integrations, was identified as the tool used. First time Google has documented AI-driven zero-day discovery by attackers at scale. (CNBC | Bloomberg | Google Cloud Blog | Fortune)

Industry & Business

  • GM lays off ~600 IT workers, immediately opens AI-skilled roles (May 11) [Hiring] — General Motors cut over 10% of its IT department — salaried workers whose expertise "no longer fits" — and is actively replacing them with AI-focused hires. This is a deliberate skills swap rather than pure cost-cutting, and mirrors Coinbase's pod restructuring (covered May 11). The pattern is becoming clear: AI-driven workforce reorgs in 2026 are targeting role profiles, not just headcount. For engineering leads, this is a signal about which skills to prioritize in hiring and which to reskill within the team. (TechCrunch)

  • Anthropic weighs $50B raise at ~$900B valuation (April 29–May 7) [Trending] — Anthropic has received multiple pre-emptive offers to raise ~$50B at an $850–900B pre-money valuation, per FT and TechCrunch; a board decision is expected in May. If priced at $900B, Anthropic would surpass OpenAI's current $852B valuation. Revenue context: Anthropic's annualized run rate reportedly surpassed $30B (up from ~$9B at end of 2025), now closer to $40B. The round may be its last before an IPO targeted as early as October 2026. (TechCrunch | TechCrunch follow-up)

Policy & Safety

  • US spy agencies seek power over AI model regulation (May 11) [Trending] — The Washington Post reports that national security officials are pushing for a larger role in AI oversight within the Trump administration, framing frontier model governance as a national security issue rather than an ethics or bias concern — a framing shift driven by AI cyber capabilities (see Security section). Separately, CAISI (Center for AI Standards and Innovation) has signed pre-deployment evaluation agreements with Google DeepMind, Microsoft, and xAI, allowing the US government to test frontier models before public release. (Washington Post | CNN Business)

Benchmarks & Evals

  • Stanford AI Index 2026: SWE-benchA benchmark that tests whether a model can resolve real GitHub issues by editing code in popular Python repositories.Jimenez et al. 2023 ↗ near 100%, model transparency down sharply (April 2026) — Key data points worth having in your back pocket: SWE-bench Verified scores rose from 60% to near 100% in a single year; frontier models now meet or exceed human baselines on PhD-level science, multimodal reasoning, and competition math. The US–China model performance gap has effectively closed. Foundation Model Transparency Index average dropped from 58 to 40 points year-over-year; documented AI incidents rose to 362 (from 233). US private AI investment hit $285.9B in 2025, 23× China's $12.4B. (Stanford HAI | Full report)

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