theSIGNAL TECHNOLOGY
14 October 2026
"Security is a process, not a product."
NEWSBUSINESSTECHNOLOGYWORLDSCIENCECULTUREENVIRONMENTSPORT
Photo: Benjamin Child / Unsplash
A television mounted in a corporate meeting room, where a hidden recording device was reportedly found.
🎙️
35
Age of the OpenAI engineer charged over the device
Discuss
  • Why might an AI company's executive meetings be considered as valuable a target as its source code or chips?
  • OpenAI says its models helped identify the culprit. How much should we trust a company's account of its own investigation?
Technology

An AI Lab Used Its Own Models to Catch a Spy

OpenAI says its models helped trace a hidden recording device to one employee, raising new questions about secrecy inside the AI race.

Somewhere behind a television in an executive conference room, a small device sat wrapped in toilet paper. It was plugged into a USB port, its contents encrypted, and it had probably been there since around 28 September. Within days, according to OpenAI, the company's own AI models had helped explain what it was and who had put it there. On Friday, federal agents arrested Abdulrahman Alfozan, a 35-year-old Saudi national and OpenAI engineer, who now faces a charge of using or attempting to use a device to intercept oral communications.

The Department of Justice says the room was used for discussions of "highly confidential technical, commercial, and strategic information". Prosecutors allege that Alfozan had no access to it, yet CCTV footage placed him nearby on several dates and showed him apparently looking behind the television. They also claim security staff later found his bag in another room, containing his wallet, his ID and a second recording device. OpenAI fired him and referred the case to the FBI.

The allegations have not been tested in court. The detail that stands out is not the hardware, which sounds crude, but the investigation. A company that sells machine intelligence reports that it turned that intelligence on a physical-security puzzle, analysing an unfamiliar device and tracing its origin. That is a striking advertisement, though it is also a claim only OpenAI has made so far, and outside investigators may want to know how much the models did compared with ordinary detective work.

//
We used our models to understand the device and where it had come from, identifying that a specific employee had planted it a few days earlier.
OpenAI, company statement
Technology

A Two-Millimeter Chip Gives Sight Back

In Sydney, a man with Stargardt disease read a single word weeks after receiving a chip smaller than a fingernail. Science Corporation's PRIMA implant replaces dead photoreceptors, while camera glasses feed it images. CEO Max Hodak argues the retina belongs to the central nervous system, so eye implants count as brain-computer interfaces.…
  • If an eye implant counts as a brain-computer interface, what else might?
Technology

Hinton Answers Nvidia: Risk Isn't Prophecy

Geoffrey Hinton says a 10–20% chance of AI-caused extinction is not a forecast but a warning. Nvidia's Jensen Huang calls the figure unscientific and hurtful. Hinton replies with a bus: told it might crash with 10% probability, nobody boards.…
  • Should unquantifiable risks like extinction be judged by evidence standards or by precaution?
AMERICAS · Technology
Rogue Anthropic AI agent gave police fake tip in unsolved murder case
GLOBAL · Technology
Nuclear war and malicious use of AI threaten end of humanity by 2100, says Lancet assessment
EUROPE · Technology
Cyber-attack similar to OpenAI Medicare breach seems likely in UK, says minister
AMERICAS · Technology
Anthropic's Claude AI submitted false murder tip to Philadelphia police
theSIGNAL IN THE LAB
1VOCABULARY
encryptedintercept
prosecutors allegeCCTV footagereferred the case
unscientificunquantifiable
2GRAMMAR FOCUS
Reported speech — statements, questions, commands with tense shift
When we report what someone said, the verb usually shifts back one tense (is → was, has found → had found, will → would), and pronouns and time words change to fit the new viewpoint. Reported questions use normal statement word order with if/whether or a question word, and reported commands use an infinitive with 'to' or 'not to'.
was · had · would · if · to · not · had been
  1. OpenAI said that the device probably been behind the television since around 28 September.
  2. Prosecutors claimed that Alfozan no access to the executive conference room.
  3. A security officer told investigators that he found a bag containing a second recording device.
  4. Reporters asked the company its AI models had really helped identify the device.
  5. OpenAI announced that it referred the case to the FBI after firing the engineer.
  6. Hinton explained that a 10% chance of a crash stop anyone from boarding a bus.
3COLLOCATIONS
Match the verb on the left with the noun phrase it best collocates with on the right.
  1. intercept
  2. refer a case
  3. face a charge
  4. replace
  5. pose a threat
4CRITICAL THINKING
Hinton's bus analogy treats a 10% risk as obviously unacceptable, yet we accept small risks of serious harm every day when we fly, drive or take medication. What makes some risks tolerable and others not, and does the analogy still work when the benefits of boarding the bus are enormous? Consider also whether a vivid image like the bus can mislead as easily as it can clarify.
5CREATIVE · HEADLINES
Write a headline for the top story in each of the following styles. One line each, no explanation:
  • TABLOID NEWSPAPER
  • LUXURY MAGAZINE
  • ACTIVIST BLOG
6WRITING
Imagine you are the Sydney patient who read a single word through the PRIMA implant. Write a short diary entry (120-150 words) describing that moment and how it changes your view of what technology can do for ordinary people.
7DEGREES OF EXTREMITY
Complete each ladder from mild to strong.
  • worrying→→
  • unusual→→
  • suspicious→→
  • a setback→→
  • valuable→→
  • doubtful→→
8SPEAKING
  1. Does a tiny implant that restores sight change what 'human' means?
  2. Would you rather your employer monitored you closely or trusted you blindly?
  3. Who should pay when a company's secrets are stolen: firms or governments?
  4. Is a dramatic warning ever more useful than a carefully hedged one?