OpenAI Moves Onto Google's Cloud, a Hacker Spends $5,000 on AI Agents, Microsoft Sells Agent Kill Switches

OpenAI's models go on sale inside Google's cloud, one hacker spends about five thousand dollars on AI agents and gets into more than a hundred companies, and Microsoft starts selling the kill switch.

On Wednesday, October 21, OpenAI said its models will be available to developers through Google's cloud, bringing its GPT products to Google Cloud's model garden and to the Antigravity coding tool in the coming weeks. OpenAI now has its models on all three of the biggest American clouds. Less than six months ago it said Microsoft would no longer hold the exclusive licence on its intellectual property, and Amazon announced plans to host OpenAI's models the day after that ended. The company was founded in 2015 partly as a counterweight to the concentration of AI research at companies like Google. Google Cloud's annualized revenue growth accelerated to about 82% in the second quarter from 63% in the first, against 37% at Amazon and 42% at Microsoft, ahead of quarterly results for all three next week. OpenAI is meanwhile courting investors at a $1.4 trillion valuation and targeting a public listing. 1

A single Chinese-speaking hacker broke into the websites of more than 100 companies, among them the farm equipment maker Deere, the appliance company SharkNinja and the engine maker Cummins, and did almost none of the work himself. Reuters reported on Monday, October 19, that the intrusion had been outsourced almost entirely to AI agents, which spent hours autonomously probing and breaking into sites. Snorre Fagerland, a principal threat researcher at the security company NorForge, watched the operation run after finding exposed instances of Cairn, an open-source AI problem-solving platform popular with hackers. The operator's instructions ran to fewer than 100 Chinese characters. From September 25 the agents went after e-commerce sites in the United States, Canada, Australia and Japan, attempting more than 1,200 in all; successful intrusions took about 23 hours of probing each. Running a mixture of China's DeepSeek and Anthropic's Claude Code, the campaign consumed roughly $5,000 of tokens, about $50 for every company it got into. Reuters could not establish how much data was taken. Deere said it had taken "necessary remediation actions" and had no indication that sensitive information was accessed; SharkNinja said the site hit carried only product support information. "It's a new world now," Fagerland said. 2

On Monday, October 19, Microsoft announced tools that treat AI agents as potential insider threats. The system keeps a continuously updated risk score for each agent, based on what data it can reach and whether its actions are starting to depart from the task it was given, and when the score rises it can strip the agent's ability to make changes while still letting it read data, require a person to approve a high-stakes step, or shut the agent down. The features ship in Microsoft 365 E7, which costs $99 per user per month, and in the company's Project Perception security system, and Microsoft is expanding the number of agent types it can detect from 25 to 60, including Claude Code, Codex and GitHub Copilot. They are in private preview, with wider availability planned for the Ignite conference in San Francisco on November 17 to 19. The approach follows chief executive Satya Nadella's argument on October 10 that controls on models have to come from outside the models: "The most trustworthy Super Intelligence system will not be the one with the model we trust most. It will be the one that enables us to trust the model the least." 3