JadidJust-in AI Developments, Incidents & Deadlines
June 2026
SMTWTFS123456789101112131415161718192021222324252627282930

AI news, Thursday, June 11, 2026

9 stories

TLDR

  • Anthropic Agent SDK billing split in 4 days (Jun 15) — automation scripts stop cold when credits run out; no rollover, no fallback unless you opt into overflow billing
  • OWASP 2026 AgenticDescribes AI systems where a model uses tools and decides its own steps instead of following a fixed script.Anthropic ↗ Security Report out today (Jun 11) — prompt injectionInput crafted to change an LLM's behavior or output in ways its builders didn't intend.OWASP ↗ attacks up 340% YoY; drives 6 of 10 agentic risk categories; real CVEsCommon Vulnerabilities and ExposuresA public catalog that gives each known security vulnerability a unique ID, a short description and at least one reference.NIST ↗ documented
  • Supabase raises $500M Series F at $10.5B (Jun 4) — AI coding agent boom (600%+ DB growth); Claude Code named the single largest contributor on the platform
  • GitHub Copilot fully on usage-based billing since Jun 1 — flat-rate request units replaced by token-based AI Credits; predictable monthly costs are now variable
  • MiniMax M3 open weightsA model whose trained weights are published for anyone to download and run.NTIA ↗ expected today — 10-day commitment since June 1 API launch; watch huggingface.co/MiniMaxAI

Action Items

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

  • Prepare for the Agent SDK billing split on June 15 — Starting June 15, Claude Agent SDK, claude -p, Claude Code GitHub Actions, and third-party Agent SDK apps draw from a separate monthly credit pool (Pro: $20, Max 5x: $100, Max 20x: $200 — billed at API rates). Credits are per-user, non-pooled, and do NOT roll over. When exhausted, automated requests stop entirely — no automatic fallback. If your team has CI/CD pipelines, scheduled agents, or background automation on Claude, either enable overflow billing now or pre-audit expected spend to avoid silent failures on June 15. Interactive Claude.ai chat and Claude Code in-terminal use are unaffected. (The New Stack | TechTimes | Zed Blog)

  • Read the OWASP 2026 Agentic Top 10 (v2.01) — Published today, maps a full year of field evidence — real CVEs, breach reports, and vendor advisories — to concrete attack patterns on agentic systems. Prompt injection (LLM01) drives 6 of 10 categories. If your team ships agents that ingest external content (emails, documents, web pages, PR descriptions), this is mandatory reading before you merge another agentic feature. (Help Net Security | DeepTeam OWASP Reference)

  • Audit Copilot team spend now — GitHub Copilot moved from flat Premium Request Units to token-based AI Credits on June 1. Plan prices are unchanged but heavy agentic usage can generate overage charges. Check your org's usage dashboard, set per-user budget caps, and note that annual plans are being retired — users on annual will roll to monthly or Free at expiry. (GitHub Changelog)

  • Migrate off Gemini 2.0 Flash if you haven't yet — Deprecated June 1; calls to the gemini-2.0-flash model string may fail or silently route. Replace with Gemini 3.5 Flash (GA since May 19, $1.50/$9 per 1M input/output tokensA small unit of data, such as part of a word, that AI models read and generate. Many AI services price by tokens in and out.NVIDIA ↗). (Google AI Pricing)

Models & Tools

  • MiniMax M3 open weights (expected Jun 11) Open Source — MiniMax committed to releasing open weights within 10 days of the June 1 API launch; that window closes today. The model leads open-weight SWE-BenchA benchmark that tests whether a model can resolve real GitHub issues by editing code in popular Python repositories.Jimenez et al. 2023 ↗ Pro at 59.0% with a 1M-token context windowHow much text a model can see and use at once while writing a reply.Anthropic ↗ and native multimodal computer-use capabilities. Watch huggingface.co/MiniMaxAI for the MiniMax-M3 card. Expect a modified-MIT license restricting commercial use without written authorization (same pattern as M2.7). (The Decoder)

  • NVIDIA Cosmos 3 (Jun 1) Open Source Robotics/IoT — NVIDIA's first open "omnimodel" for physical AI: a two-tower Mixture-of-Transformers combining a vision-language reasoning tower with a diffusion-based world/action generator. Trained on 20T multimodal tokens. Ships as Cosmos 3 Nano (16B, workstation/edge robotics) and Cosmos 3 Super (64B, datacenter). NVIDIA open-sourced 6 synthetic datasets covering robotics, driving, physics simulation, and warehouse environments alongside training scripts and inferenceRunning a trained model on new input to get an output, as opposed to training it.NVIDIA ↗ tooling. (NVIDIA Newsroom | NVIDIA Technical Blog)

  • OpenAI + Oracle Cloud: models and Codex via cloud commitment (Jun 10) — OpenAI announced that Oracle customers can apply existing Oracle Universal Credits toward OpenAI models and Codex through OCI. Access is rolling out over the coming weeks. Part of OpenAI's broader push to distribute beyond its primary Microsoft arrangement ahead of its September IPO. (OpenAI)

APIs & Pricing

  • Anthropic: Agent SDK credit pool split (Jun 15 deadline) Trending — See Action Items above. The key implication: any team treating Claude Pro/Max as a flat-rate API-equivalent for agentic workloads will hit a hard wall on June 15 unless they've set up overflow billing or migrated to direct API access. (The New Stack)

  • GitHub Copilot: AI Credits replace request units (Jun 1) Trending — All plans now billed on token consumption. Annual plans retired. Copilot code review now also consumes GitHub Actions minutes in addition to AI Credits. Per-user budget controls available in org settings. (GitHub Blog)

  • Gemini 2.0 Flash deprecated (Jun 1) — Removed from active support. Stable migration target: Gemini 3.5 Flash (GA, $1.50/$9 per 1M input/output). Gemini 3.5 Pro remains in limited Vertex AI preview with no GA date confirmed. (Google AI)

Security

  • OWASP 2026 Agentic Security Top 10 (v2.01) (Jun 11) Trending — OWASP's updated report catalogs prompt injection as the single fastest-growing AI attack vector (340% YoY) and maps it to 6 of 10 categories in their Agentic Applications Top 10. Documented real-world CVEs: CVE-2025-53773 (RCERemote code executionA flaw that lets an attacker run malicious code on a target's computers or network.Cloudflare ↗ via injected PR descriptions in GitHub Copilot, 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 ↗ 9.6), CVE-2026-22708 (Cursor IDE environment poisoning, CVSS 9.8), and a June 2025 zero-clickAn attack that compromises a device without the user clicking, opening or interacting with anything.F5 ↗ Copilot incident extracting OneDrive/Teams data (CVSS 9.3). Core message: in agentic systems there is no reliable way to mark retrieved tokens as "data" vs. "commands" — guardrails must go at the tool/action layer, not just the prompt layer. (Help Net Security | OWASP GenAI Project)

Industry & Business

  • Supabase raises $500M Series F at $10.5B valuation (Jun 4) Infra — The open-source Postgres backend doubled its valuation in 8 months on the back of AI coding tool adoption. Database launches grew 600%+ in the past year; 60%+ are triggered by AI tools, with Claude Code named the single largest contributor on the platform. Led by GIC, with Stripe (increased position), Salesforce Ventures (new), Accel, and Y Combinator participating. Total capital raised now exceeds $1B. (CNBC | TechCrunch)

Policy & Safety

  • Colorado AI Act replaced: compliance deadline moves to Jan 1, 2027 (May 14, signed) — Governor Polis signed SB 189, replacing the original Colorado AI Act with a narrower law. Gone: the duty-of-care risk management framework, mandatory algorithmic impact assessments, and AG reporting obligations. What remains: disclosure requirements around automated decision-making. A federal court also separately stayed enforcement of the original law in April. If your team was preparing for the original June 30, 2026 deadline, the burden is now lighter and the timeline extended — but watch for AG rulemaking before January 1. (Hunton Privacy Blog | Alston & Bird)

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