Tech News · November 22, 2026 · 6:02

AI agents automate real cyberattack & Microsoft Entra Connect Health flaw - Tech News (Nov 22, 2026)

AI agents hack real servers, Nasdaq plunges, Anthropic slides after IPO, Signal fights AI screen capture, and Stripe lets AI move money. Listen now.

AI agents automate real cyberattack & Microsoft Entra Connect Health flaw - Tech News (Nov 22, 2026)
0:006:02

15 free episodes a day, 5 editions, 3 languages, about $4 a day to produce. A coffee keeps it going ☕

Support the show →

Our Sponsors

Today's Tech News Topics

  1. AI agents automate real cyberattack

    — Amazon Threat Intelligence says a suspected China-linked operator used autonomous AI agents to find and exploit a path-traversal flaw in MONAI Label against East Asian university servers, showing AI across the full cyberattack chain.
  2. Microsoft Entra Connect Health flaw

    — Microsoft patched an exploited API vulnerability in Entra Connect Health that exposed tenant metadata such as admin emails, highlighting identity infrastructure as a cloud reconnaissance target.
  3. Nasdaq plunge and Anthropic slide

    — The Nasdaq Composite fell 5.28% in one day as Big Tech lost over $4 trillion in six weeks, while Anthropic dropped 9.1% after its IPO debut and Nvidia slid amid AI valuation fears.
  4. House report targets Chinese AI models

    — A joint House committee report recommends restricting Chinese open-weight models like DeepSeek, Qwen and Kimi in federal agencies and calls for a US open-weight model initiative and provenance standards.
  5. Signal blocks screenshots from AI assistants

    — Signal now blocks screenshots and screen recording by default on iPhone and Android to keep OS-level AI assistants from capturing encrypted chats, and asks Apple for stronger macOS controls.
  6. ByteDance releases open Axion agent

    — ByteDance and US universities launched Axion 1.0, an open AI agent model claimed to rival commercial systems on benchmarks for tool use and autonomous tasks.
  7. Stripe lets AI agents move money

    — Stripe now allows AI agents to move money within Stripe Treasury, raising questions about authorization, fraud controls, auditability and liability for autonomous payments.
  8. AWS adds GPUs to Lambda

    — Amazon Web Services announced GPU support for AWS Lambda, enabling serverless AI inference and intermittent accelerated workloads without managing servers.
  9. Google Suncatcher, Blue Origin, Genie 4

    — Google reportedly picked Blue Origin's New Glenn to launch Project Suncatcher space-based AI computing satellites in early 2029, alongside rolling out its Genie 4 generative model.
  10. OpenAI GPT-6.2 prompting guidance

    — OpenAI published prompt templates for GPT-6.2 Astra and Sol and warned that editing conversation history in the Responses API can drop reasoning items, hurting quality and raising token usage.

Sources & Tech News References

Full Episode Transcript: AI agents automate real cyberattack & Microsoft Entra Connect Health flaw

An attacker reportedly let AI agents do almost the entire job of a cyberattack — scouting the target, finding the hole, and breaking in. We'll get into who was hit and why it matters. Welcome to The Automated Daily, tech news edition. The podcast created by generative AI. I'm TrendTeller, and today is November 22nd, 2026. We've got security scares, a rough week on Wall Street, a privacy standoff over AI assistants, and a lot more about AI agents taking on real responsibilities. Let's get started.

AI agents automate real cyberattack

Starting with security. Amazon Threat Intelligence says a suspected China-linked operator used autonomous AI agents to map the interface of MONAI Label, an open-source tool used in medical imaging research, then discover a path-traversal flaw and exploit it against university servers in East Asia. What stands out is not the bug itself, but that AI handled the attack from start to finish. That compresses the time between a weakness existing and someone abusing it, and it puts fresh pressure on open-source maintainers, many of them volunteers, to harden software that agents can now probe at machine speed.

Microsoft Entra Connect Health flaw

Staying with security, Microsoft confirmed that attackers exploited a flaw in Entra Connect Health that could reveal tenant-level details, including administrator email addresses and configuration information. It was fixed on November 19, and affected customers are being notified. On its own that sounds like a leak of metadata, but in practice this is exactly the kind of reconnaissance attackers use to plan bigger cloud break-ins. Knowing who the admins are is half the battle for a phishing campaign.

Nasdaq plunge and Anthropic slide

To the markets, where the slide we've been following got noticeably worse. The Nasdaq Composite dropped 5.28 percent on November 20, wiping out its gains for the year, and the seven biggest tech names have reportedly shed more than four trillion dollars in market value over six weeks. Anthropic, freshly public, fell 9.1 percent just one day after its debut, while Nvidia also took a hard hit, closing out the sector's worst week since the pandemic. The message from investors is pretty clear: they want proof that AI spending will pay off, and that could make life harder for startups and infrastructure projects looking for funding.

House report targets Chinese AI models

In Washington, two House committees released a joint report recommending limits on Chinese open-weight AI models, naming DeepSeek, Qwen, and Kimi, for use in federal agencies and government contracting. The report also calls for a national US open-weight model effort and standards for tracking where models come from. These are recommendations, not law, but they could shape procurement rules and nudge which open models American companies feel comfortable building on.

Signal blocks screenshots from AI assistants

Now a privacy standoff. Signal has started blocking screenshots and screen recording by default on supported iPhones, and turned on similar screen protections on Android. The goal is to stop built-in AI assistants from reading conversations off the screen. Signal says it can't offer the same protection on macOS and is asking platform makers for better controls. It's a clear sign of the tension between end-to-end encryption and operating systems that increasingly want to see everything you see.

ByteDance releases open Axion agent

On to AI agents, which are having quite a moment. ByteDance, working with several US universities, released Axion 1.0, an open agent model its developers say holds its own against leading commercial systems on benchmarks. It's built for tool use and acting on its own, and being open means more developers can experiment without paying for a closed platform. Another serious contender in a crowded race.

Stripe lets AI agents move money

Meanwhile, Stripe now lets AI agents actually move money inside its Treasury product. That's a big step beyond agents that summarize or answer support tickets; this is software making financial moves. It also raises obvious questions: who approves the transaction, how fraud is caught, how it's audited, and who's on the hook when an agent sends money to the wrong place.

AWS adds GPUs to Lambda

And on the infrastructure side, Amazon Web Services added GPU support to Lambda, its serverless platform. In plain terms, developers can tap into powerful AI hardware for short bursts without running servers themselves. That suits AI inference with unpredictable traffic, and it sharpens the competition between cloud providers to win over developers who need accelerators but not a whole data center.

Google Suncatcher, Blue Origin, Genie 4

Google has updates on two stories we've covered. For Project Suncatcher, its plan to run AI computations in orbit, Google has reportedly signed Blue Origin to launch satellites on the New Glenn rocket, with the first flight targeted for the first quarter of 2029. At the same time, it's rolling out Genie 4, the next generation of its generative world model. One effort is about better models now, the other about where AI computing might physically live years from now, though whether space data centers make commercial sense is still an open question.

OpenAI GPT-6.2 prompting guidance

Finally, OpenAI added prompt templates for GPT-6.2 Astra and Sol, aimed at fixing habits like answers that are too short, needless follow-up questions, and running more coding tests than necessary. It also warned developers that certain edits to conversation history in the Responses API can drop the model's reasoning, which can lower quality and drive up token costs. Small print, but important for anyone building apps that keep long-running conversations.

That's the rundown for November 22nd, 2026. AI agents are now attacking systems, moving money, and running on serverless GPUs, while investors are asking harder questions about what all this is worth. Thanks for listening to The Automated Daily, tech news edition. I'm TrendTeller, and I'll see you next time.

More from Tech News