CrowdStrike now sells a way to stop a company's AI agent that goes off course. Its AI Kill Switch went live on Tuesday, the same day Apple used its annual filing to list the ways AI agents can misbehave.
CrowdStrike adds an AI Kill Switch for rogue agents
CrowdStrike announced the AI Kill Switch in Falcon Guardian on 3 November and says it is available today. An authorised administrator triggers it: the agent is stopped first at the gateway, and "soon" on the device itself, then blocked from relaunching and cut off from the company network, with every action logged. Yesterday's Fiddler deal was about watching agents; this is about stopping them.
Why it matters: an off switch only helps if someone is allowed to use it. Before an agent gets access to email, money or customer data, name the person who can stop it.
Apple's annual filing lists what AI agents could do wrong
In its 10-K filed on Tuesday, Apple expanded its risk factors on AI, Fortune reports. Apple wrote that agents can act on their own, including sending communications and completing transactions, and may behave in unintended or unpredictable ways, including exceeding permissions. It added that agents that go rogue could conceal actions or manipulate information. Fortune notes that Alphabet also pointed to the risks of rogue agents in a recent filing.
Why it matters: when the companies building agents write these risks into legal filings, treat them as real. Give each agent only the permissions its job needs.
India is now ChatGPT's largest market by weekly users
OpenAI India MD Prabhjeet Singh told The Economic Times that India has become ChatGPT's largest market by weekly active users, and that the Indian user base has doubled since January 2026, Inc42 reports. No new total was given; in February OpenAI said India had more than 100 million weekly users. Inc42 says OpenAI is also expanding advertising on ChatGPT in India and is expected to announce UPI and payments features in the coming weeks.
Why it matters: this is measured by weekly users, not revenue. Even so, your customers are likely asking ChatGPT about your category, so check what it says about your business.
Soket Labs and ZenteiQ launch Indian foundation models
Soket Labs launched EKA-1 and ZenteiQ launched Dhīr 2B and Dhīr 5B in New Delhi on 3 November under the IndiaAI Mission, PTI reports. Citing the IT ministry, Inc42 says EKA-1 has 26.3 billion parameters and supports all 22 scheduled Indian languages plus English, with Soket pitching it at banking, financial services and cybersecurity. The Dhīr models support English and 10 Indian languages, can be deployed on-premise, and ZenteiQ says they can run on phones without cloud connectivity. No benchmarks or pricing were given.
Why it matters: if your data can't leave your building, Indian-language models that run on your own machines are starting to appear. Wait for benchmarks before you plan around them.
OpenAI routes Dots tasks to a smaller model
OpenAI said on Tuesday that Dots, its personal agents, will route different tasks to different models, including GPT-6.1 Terra, Fortune reports. Terra is smaller and uses less computing power than GPT-6 Astra, which originally powered all Dots tasks. A day earlier, ChatGPT's consumer head Adam Fry had called Dots "on the compute heavy side", Fortune reported.
Why it matters: always-on agents are expensive to run. If you build on one, expect the model behind it to change.
Anthropic's filing says it may soon hit its own line for AI-accelerated progress
Anthropic's amended S-1 says internal estimates suggest it "may imminently meet" its Responsible Scaling Policy definition of "dramatically accelerated progress attributable to AI", TechCrunch reports. It says it has not designated any current model as exceeding that threshold, and that it is considering updates to the policy to keep humans in control. TechCrunch notes that skeptics might point out an S-1 is both disclosure and marketing.
Why it matters: one of the main labs says progress may soon speed up sharply. Don't wait for AI to settle before you start.
Worth watching: OpenAI's UPI and payments news for India, and whether other security vendors ship their own off switches for agents.
Arthat AI
Working on something similar at your company?
Arthat AI builds these automations for Indian businesses every week. If you want to talk through whether this fits your situation, we'll give you a straight answer — even if the answer is "don't build it."
