OpenAI Engineer Arrested Over a Hidden Recorder in an Executive Conference Room

Federal prosecutors say Abdulrahman Alfozan, a 35-year-old OpenAI engineer, hid an audio recording device in an executive conference room at the company’s San Francisco headquarters. According to the San Francisco Standard’s report on the criminal complaint, the device was plugged into a USB port on the back of a television and concealed in toilet paper. OpenAI staff found it on September 30, and it was determined to have been installed around September 28. Its contents were encrypted.

Prosecutors allege a second, similar device was found in an unattended bag containing Alfozan’s wallet and ID on October 5. He was arrested last Friday and charged with using or attempting to use a device to intercept oral communications. He faces up to five years in prison and a $250,000 fine if convicted, and a detention hearing is set for October 19. Alfozan has not been convicted, and these are allegations from a complaint.

OpenAI says it has no indication that model weights, source code or customer data were accessed. The company says it fired him the day he was arrested and that investigators used its own AI models to analyze the device. AI Weekly, citing Bloomberg, reports the board was a roughly $14 Seeed Studio dev board. The case lands as OpenAI pushes always-on agents into more workplaces, like the one in our Dots by OpenAI review.

Google’s Gemini 4 Argon: $2 per Million Input Tokens and a Lead on DeepSWE

Google announced Gemini 4 Argon on September 30, and its blog post sets introductory pricing at $2 per million input tokens and $10 per million output tokens, with cached input 95% cheaper. After the introductory period the price rises to $4 and $20. Google claims a state-of-the-art 77.9% on DeepSWE v1.1, 91.7% on the LVBench long-video test, a tie for first at 68% on CWE-bench for fixing security vulnerabilities, and the top spot on Zapier’s AutomationBench at 51.3%. These are Google’s own numbers.

The model’s output limit rises to 1 million tokens, up from 64K. Google says Argon started with trusted cyber defenders through its Fairwind Program, and that wider access will begin with paid API customers and Google AI Ultra subscribers. AI Weekly reports Google began a broader rollout to AI Pro and Ultra subscribers on October 13. I couldn’t independently confirm that date, so check the Gemini app for your account.

For developers comparing coding agents, see how Cursor handles agent work in our Cursor AI review.

Tencent Raises $5.7 Billion in Bonds Mostly to Fund AI Spending

Tencent raised about $5.7 billion through a four-part debt offering, according to Reuters as summarized by AI Weekly. It sold $1.3 billion of 10-year and $500 million of 20-year dollar bonds, plus 20 billion yuan of 10-year and 6 billion yuan of 30-year offshore yuan bonds. The order book topped $23 billion, more than four times the final size and well above an initial $4.5 billion target.

Tencent told investors the proceeds will mainly fund AI capex as it ramps data-center and model spending while US chip restrictions continue. The demand is another sign that AI infrastructure is increasingly being financed with debt, a theme we also covered in our social posts on SpaceX’s reported $40 billion Nvidia chip financing.

Apple Skips the EU for Its New HomeView Because of Siri AI and the AI Act

At its Welcome Home event, Apple unveiled the $299 HomeView, a 7-inch smart display with face recognition for up to six users, along with a $149 HomePod mini 2 and a $299 fourth-generation Apple TV 4K, according to MacRumors via AI Weekly. Pre-orders opened October 13, and the products ship November 13 in eight countries.

HomeView will not launch in the EU. Apple says the advanced Siri AI capabilities it requires aren’t available there under the EU AI Act, and it says it is working constructively with the EU without giving a timeline. It’s a concrete example of AI regulation changing which products reach European customers.

Microsoft Patches More Than 1,400 Security Flaws and Credits AI for the Record

Microsoft’s October patch release covers more than 1,400 security flaws, according to KrebsOnSecurity as summarized by AI Weekly. That beats September’s record of 972 and is roughly ten times a normal month. At least 78 are rated critical, and one Windows Shell bypass, CVE-2026-96951, is already being exploited in the wild.

Microsoft credits large language models for the surge in bug discovery and remediation. The company has now shipped more than 4,000 CVE fixes in 2026, passing the 2020 record of 1,261 with two months left. The practical takeaway for IT teams is to expect larger patch queues as AI finds more bugs faster than people can review them.