Skip to main content

Critical GitLab RCE, Meta's frontier model and Microsoft's chip bet

  1. Security

    GitLab patches CVSS 10.0 unauthenticated RCE in its CI/CD regex engine (opens in a new tab)

    GitLab shipped 19.4.2, 19.3.5 and 19.2.8 to fix CVE-2026-97229, a CVSS 10.0 flaw letting unauthenticated attackers run code via a crafted regex in CI/CD config. Self-managed instances should upgrade now; GitLab.com is already patched.

    Source: GitLab Docs

  2. AI

    Meta launches Muse Wonder, its first frontier model, through a paid API (opens in a new tab)

    Meta released Muse Wonder and Muse Wonder Full, its first frontier-class models, via the Meta Model API at $4/$20 per million input/output tokens. Builders get another closed, paid option for agentic and coding work; benchmark parity claims are Meta's own.

    Source: TechCrunch

  3. Cloud

    Microsoft unveils Maia 300 chip, will make Maia its main AI inference system (opens in a new tab)

    Microsoft unveiled the TSMC N2P Maia 300; cloud chief Scott Guthrie expects 40% more tokens per megawatt than leading GPU systems (comparison systems unnamed) and Maia to be Microsoft's primary inference system from 2027. Azure AI economics will lean less on Nvidia.

    Source: CNBC

  4. Security

    Hackers hijack legacy Danfoss controllers to heat fridges in French supermarkets (opens in a new tab)

    Attackers abused legacy Danfoss AK-SM800 and AK-SC355 refrigeration controllers, which no longer get security updates, to heat fridges at 300 Intermarché/Netto and under 100 U stores, per BFMTV, spoiling stock. Engineers should segment, monitor and retire unpatched OT devices.

    Source: NBC News

  5. Business

    Character.AI files for Chapter 11, agrees to sell some assets to Disney for $20M (opens in a new tab)

    Character.AI filed for Chapter 11 in Delaware, with Disney as stalking-horse bidder paying $20M for a non-exclusive IP licence. Its filing says AI entertainment "has proven to be costly to develop and grow", a warning for consumer AI apps.

    Source: TechCrunch

  6. Business

    HCLTech signs Anthropic partnership, sets up dedicated Claude business unit (opens in a new tab)

    India's HCLTech signed a Global Premier Partnership with Anthropic, creating a dedicated business unit and embedding Claude into HCL BigFix, AppScan and its AI Force platform. Like Accenture's Anthropic Business Group, IT services giants are rebuilding delivery around agentic AI.

    Source: DigiTimes

  7. Regulation

    EU opens consultation on how AI Act penalties will be enforced (opens in a new tab)

    The European Commission opened a consultation, running to 9 November, on a draft implementing act setting procedures for AI Act penalties, including company safeguards and limitation periods. AI providers facing EU enforcement should review the draft and comment.

    Source: Agence Europe

My take

Today's thread is that the AI stack is getting more expensive to run and more dangerous to leave unpatched. Meta and Microsoft are both betting on owning more of it (models and silicon), while Character.AI shows what happens to apps that rent it at scale. Meanwhile GitLab's CVSS 10 bug and the French fridge hack are reminders that the boring infrastructure underneath still decides whether you have a bad week, and Brussels is now writing the rulebook for how AI fines actually get levied.