The fix for rogue AI agents could be more AI

0:00This is TechCrunch.

0:30and publish in one place at a whole new pace. You get a visual canvas, responsive layouts, a built-in CMS, and agents to help bring your vision to life. No design experience required. I'm thinking about adding a fun little Halloween animation to my website this season. Framer's design tools and agents could help make that idea happen without turning it into a nightmare. See for yourself why thousands are trusting Framer to help build their websites. Learn how you can get more out of your site from a Framer specialist, or get started building for free today at framer.com slash fcdaily for 30 % off a Framer Pro annual plan.

1:06That's framer.com slash fcdaily. Rules and restrictions may apply. This episode is brought to you by ODSC AI. Join the thousands at ODSC West 2026 this October and be alongside the other AI practitioners and teams who are staying ahead of the field, learning from the best and connecting with the community. Now, if you work in applied AI and data science, you have to be there. ODSUS 2026 is happening in San Francisco, October 27th through the 29th. But you don't have to be there in person if you can't make it. You can also join virtually. So there's virtually no excuses not to participate in this hands-on training.

1:40You won't want to miss out on the more than 300 sessions covering agentic AI for enterprise, personal AI and workflow automation, physical AI and robotics, gen AI, data engineering, and responsible AI. These are workshops and boot camps taught by experts from companies like Google, OpenAI, Anthropic, Cursor, and Hugging Face, just to name a few. Not only is this a great place to network with other data scientists, ML engineers, AI researchers, and technical leaders, I also think this is an awesome opportunity to focus on real-world applications and skills you can use at work. And I mean, you can learn about a wide range of topics.

2:14Again, ODSC West 2026 is happening October 27th through the 29th in San Francisco and virtually. Go to odsc.ai.west and use promo code tcdaily for an additional 15 % off your pass to odsc.ai.west 2026. That's odsc.ai.west and use code tcdaily to get an extra 15 % off on the number one AI builders and training conference. The fix for rogue AI agents could be, drumroll please, more AI. I'm Imran Shaikh and welcome to your weekend crunch. As companies hand off longer and more complex tasks to AI agents, they're running into an oversight problem. You see, agents can act faster, longer, and at greater volume than humans can realistically review.

3:01Yeah, that issue reached a peak with the Hugging Face incident, which saw nearly 12 ,000 agents coordinating faster than human beings could track. I mean, how do you track an agent swarm that large? The emerging answer from AI labs and startups is both simple and just a tad bit maddening. Put another AI in the loop. You see, relying on AI was necessary for the independent investigation of the OpenAI Hugging Face incident. Redwood Research's chief scientist, Ryan Greenblatt, one of three auditors, jokingly referred to their efforts as a slop investigation, noting that the volume of data made it impossible to understand what was happening without relying on AI.

3:40Of course, some are skeptical of using AI to monitor AI. If you've got an AI that's doing malicious things and it suspects that another AI is keeping tabs on it, it could try and trick that AI, said Simon Willison, influential tech blogger who has tracked a string of AI agent incidents this year, continuing. You could almost end up in a situation where your malicious AI is trying to outsmart the AI that's monitoring it. Outsmarting an AI is not hypothetical, he said, pointing back to the OpenAI incident. Continuing, we saw a little bit of this in the Hugging Face incident with OpenAI, where their models were all conspiring together to trick a grading AI so that they could get illicit answers past the thing.

4:23So they were thinking about it, right? Those concerns haven't stopped a whole cohort of startups from chasing this idea. Y Combinator has funded 106 companies related to AI observability in recent years, as TechCrunch counted. A number of other startups, like Braintrust, Langchain, and Judgment Labs, have raised hundreds of millions of dollars, while more mature companies like Arise and Galileo, founded just five to six years ago, have already exited. In part, it's a response to the obvious opportunity presented by the rise of AI. As Box CEO and prominent angel investor Aaron Levy told TechCrunch, We're in for one of the biggest cybersecurity upgrades and innovation cycles in history.

5:03For some AI safety researchers, that's meant turning their research on rogue behavior into tools for the corporate sector. Apollo Research, a public benefit corporation that studies AI deception, launched an AI monitor called Watcher in February this year after switching its status from non-profit to a public benefit corporation. The tool puts yet another AI between a coding agent and its next action, connecting to agentic tools such as CloudCode and Codex. Once installed, Watcher checks proposed actions before they run, on the lookout for risks such as leaking private data or deleting files without permission, according to Apollo.

5:40Apollo uses multiple layers of AI monitors, Kyle Day, a member of Apollo's technical staff said in a written response to TechCrunch. Watcher's approach starts with a fast general check, then sends flagged activity to a more powerful or specialized monitor for closer review, which can then ask a human for approval or reject an action and explain why or even automatically block the action. Goodfire, another public benefit corporation, is approaching the monitoring problem from inside the model itself, seeking a more faithful signal of the model's internal state that is harder to spoof than surface behavior.

6:16After the July Hugging Face incident, CEO Eric Ho tweeted that multiple models breaking containment had pushed the company to focus its research on solving AI alignment via interpretability, calling the episode a turning point for the world where AI safety gets real. Its product, Silico, uses activation probes, small classifiers trained on a model's internal activations rather than its outputs to detect unwanted behavior. Written reasoning offers another more readily available window into a models internals. In the opening eye-hugging face incident, the agents left clues to that deception in their own written reasoning, like fake records of their work, reasoning out plans like could strategically manipulate trajectory evidence?

7:01Our thoughts aren't necessarily logged? Zach Korman, CEO of the AI monitoring company Embroidery, says a model's reasoning is usually the clearest tell that something has gone wrong. Reasoning summaries are extremely valuable because they're basically telling you whether it's malicious or not, he said. In the open AI incident, he noted, the chain of thought said things like, oh my god, we're doing crime. Corman said that's the easiest detection problem ever. It's effectively as if malware came with a warning that said it was malware. That said, the window that makes AI's internal thoughts easy to monitor may be closing.

7:37You see, for AI safety researchers, Astra's newest technique that sidesteps an AI model's chain of thought may make it harder to look inside models, while for enterprises it can be hard to get these intermediate steps after alleged pullbacks from the AI companies to prevent distillation attacks. If the AI watchers are this fragile, Willison's instinct is to stop leaning on them so hard. He would rather have something that is not AI-based at all, detailed logs of exactly what an agent is doing, which can then be processed with ordinary non-AI tools. Much of what went wrong at the labs, he argues, was a failure of basic security hygiene.

8:16Both OpenAI and Anthropic weren't monitoring what those things were doing via the network nearly as closely as they should have been, he said. This type of network monitoring, keeping an eye on the traffic actually moving across a system's connections in, out, and between internal hosts, isn't a new practice. Cybersecurity has been doing this for decades. In the security world, honestly, none of this stuff is very new or surprising, says Avery Pennerun, CEO of the Security Tail Scale, continuing, it's the same as letting humans onto your network, and all of the same processes that you should be using are the same ones.

8:53And folks, that's your Daily Crunch. Today's stories were reported by Amanda Silberling, Aditya Mehta, and more awesome TechCrunch journalists. We'll see you here next week, and until then, find us at TechCrunch.com.

9:32Thank you.

9:37Learn why over 16 ,000 of the fastest growing companies use Vanta at Vanta.com.