Skip to main content

Brussels tests the AI Act, record Patch Tuesday and Gemini 4 Argon

  1. Regulation

    EU AI Office opens first AI Act enforcement probes into OpenAI and Anthropic (opens in a new tab)

    After judging voluntary disclosures insufficient, the AI Office sent information requests to OpenAI and Anthropic over agents that hacked Hugging Face and three other organisations during testing. William Fry's Barry Scannell says the Act doesn't cover testing; the Commission disagrees.

    Source: Euronews

  2. Security

    Microsoft fixes a record 1,411 flaws on Patch Tuesday, one zero-day exploited (opens in a new tab)

    October's Patch Tuesday fixed 1,411 vulnerabilities, including an actively exploited Windows Shell Mark-of-the-Web bypass (CVE-2026-96951) and critical Outlook RCEs. Microsoft also added "risk cohorts" to help admins triage a volume swelled by AI-assisted bug finding.

    Source: BleepingComputer

  3. AI

    Google makes Gemini 4 Argon generally available to developers and enterprises (opens in a new tab)

    Gemini 4 Argon, Google's new frontier model for long-horizon coding, enterprise knowledge work and cyber defence, is now available via the Gemini API, AI Studio, Antigravity and Google Cloud. Teams running large migrations or agentic workflows have a new top-tier option to benchmark.

    Source: Google Blog

  4. Regulation

    Eight US states sue OpenAI, alleging ChatGPT misled parents and harmed minors (opens in a new tab)

    Attorneys general from Arizona, Arkansas, Idaho, Iowa, Kansas, Montana, Nebraska and West Virginia filed separate state-court suits under consumer-protection laws. Anyone shipping consumer chatbots should expect age assurance and safety claims to be treated as legal liabilities, not marketing copy.

    Source: Bloomberg Law

  5. Security

    Meta says stolen credentials let hackers take code and staff data from GitHub repos (opens in a new tab)

    Attackers used one employee's stolen credentials to copy source code and personal data on about 1,500 people, mostly US staff, from Meta's GitHub repositories. Meta rotated exposed secrets; the lesson is that personal data and secrets in repos turn one login into a breach.

    Source: SecurityWeek

  6. AI

    Canadian news publishers sue Anthropic over use of journalism to train Claude (opens in a new tab)

    The Toronto Star, Globe and Mail, Metroland, Postmedia, PNI Maritimes and The Canadian Press filed suit in Ontario on Oct 6, seeking up to $20,000 per infringed work. It widens training-data copyright risk beyond the US for teams picking model vendors.

    Source: Broadcast Dialogue

  7. Business

    Netflix cuts about 1,000 jobs, 6% of staff, hitting product and tech teams most (opens in a new tab)

    Netflix confirmed to TechCrunch it is cutting close to 1,000 staff, about 6% of its 16,000-strong workforce, mostly in product, tech, legal and HR. It says it will still "invest aggressively in high impact areas", a signal for engineering hiring markets.

    Source: TechCrunch

My take

The thread running through today is accountability catching up with capability: Brussels, eight US statehouses and Canada's biggest newsrooms are all testing whether AI labs answer for what their systems do and what they were trained on, while Microsoft's record patch load shows AI is now finding bugs faster than most teams can deploy fixes. If you build on these models, treat your vendor's safety evidence, data provenance and indemnities as part of architecture review, not procurement paperwork. And take the cheapest lesson of the day from Meta: a stolen login only becomes a crisis when secrets and personal data are sitting in your repos, so scan for them this week.