Skip to main content

Trust breaks at the seams: OAuth apps, stored keys and AI answers

  1. Security

    ConnectWise suspends Cloud Backup after stolen app credential hits Microsoft 365 tenants (opens in a new tab)

    ConnectWise has switched off its Cloud Backup service after attackers used a stolen app credential to sign in to customers' Microsoft 365 tenants. If any vendor app holds broad Graph permissions in your tenant, review its access and sign-in logs now.

    Source: Help Net Security

  2. Security

    OpenCode Console breach may have exposed users' OpenAI, Anthropic and Google API keys (opens in a new tab)

    An intruder reached a legacy OpenCode Console system, putting provider API keys saved by 1.4% of Console users at risk. Deleting a key in Console does not revoke it, so affected users should rotate keys directly with each provider.

    Source: The Hacker News

  3. Regulation

    Second German court holds Google liable for false claims in its AI search answers (opens in a new tab)

    In a September 24 injunction, Aachen's court followed Munich's reasoning and treated Google's AI Mode answers as Google's own statements, after the AI wrongly tied a doctor to another man's crimes. Teams shipping AI answers in Europe should treat generated text as their own words.

    Source: The Decoder

  4. Business

    Anthropic puts 5% of its shares into a new Public Benefit Fund (opens in a new tab)

    If stockholders approve, Anthropic will move 5% of its equity into a fund that could help with AI's impact on jobs, and the board can add more as the company's value grows. It shows how AI labs want to look as automation changes engineering work.

    Source: CNBC

  5. Security

    Unpatched Broadcom Wi-Fi bug lets nearby attackers get root on millions of routers (opens in a new tab)

    Researcher Quentin Kaiser showed that a flaw in Broadcom's WPS setup code gives root access without the Wi-Fi password, including gateways from Proximus, Orange Belgium and VOO. There is no CVE or fix yet, so check whether WPS can be turned off on your devices.

    Source: The Hacker News

  6. Cloud

    Microsoft pulls out as anchor tenant of Nscale's Essex AI data centre (opens in a new tab)

    Microsoft has walked away from Nscale's planned Loughton site, which reportedly may not get grid power until the 2030s, as Nscale prepares to raise up to $3 billion in an IPO. It shows power, not GPUs, is now the real limit on Europe's AI capacity.

    Source: The Next Web

  7. Security

    Hackers exploit SiteGuard WAF flaw to run commands as root or SYSTEM (opens in a new tab)

    Attackers are abusing CVE-2026-106931 in the statistics feature of Japan's SiteGuard Server Edition firewall, which is on by default. Patch now, but save diagnostic data first, because the vendor warns that updating can wipe signs of an attack.

    Source: BleepingComputer

  8. Business

    Disney agrees to hire most of bankrupt Character.AI's workforce (opens in a new tab)

    A court filing shows Disney offering jobs to at least 60 Character.AI staff, worth an estimated $60 million a year in pay, next to a $20 million bid for some assets. Another bidder could still top it, but AI talent is clearly the prize.

    Source: TechCrunch

My take

Today's biggest stories share one lesson: things break where we hand over trust. A backup app with wide Microsoft 365 rights, a web console holding other companies' API keys, a router running Broadcom code that dates back to at least 2010, and an AI answer box that users are told to double-check all failed at that handoff. If you build software, treat every secret you store and every sentence your AI writes as something you own, because attackers and German judges already do.