14 October 2026
The Biggest AI Security Risk Is Still a Human Insider
The most serious security story at an AI lab this month wasn’t about an AI. It was an allegation about a person, a badge and a hidden recorder.
Key takeaways
- Federal prosecutors have charged an OpenAI engineer over an allegedly hidden office recorder
- Most founders spend heavily on AI security and very little on insider controls
- A dedicated sweep team found the device within about two days of it going in
- Access, physical and people controls deserve the same attention as model and data safeguards
- AI usage logs can become evidence, so decide what you keep and who reviews it
I’ve spent years around fast-growing companies, first at Obby and Baluu and now through FirstMotion and my advisory work at VAQA. In almost every board pack I’ve seen, the AI risk slide is getting longer while the people and access slide stays the same size.
This story is a useful reminder of why that balance is off, so here’s what was reported and what I’d do about it.
What was reported
According to The Rundown AI, US federal prosecutors have charged an OpenAI engineer with trying to eavesdrop on company leaders using a recording device hidden in the San Francisco office. OpenAI’s bug-sweeping team reportedly found the recorder on 30 September, tucked inside toilet paper behind a meeting room TV, about two days after it allegedly went in.
Prosecutors also say a chatbot account linked to the engineer’s spouse asked how to listen in remotely, how to encrypt a memory card and how to hide a homemade recorder. OpenAI says investigators fed forensic copies of the device’s storage to its own models, which helped explain the hard-to-detect hardware and point to the employee.
These are allegations, and the engineer is entitled to the presumption of innocence. The case hasn’t been tied to any foreign government.
Why this matters if you don’t run an AI lab
OpenAI has a team whose job is to sweep for bugs. Almost no company with 20 or 200 people does.
That gap is the point. The attack here, if the allegations hold, didn’t need a clever exploit. It needed someone trusted who knew where the meetings happened and could walk in with a badge.
Over the past two years, most security conversations I’ve sat in have been rewritten around AI: prompt injection, data leakage into models, agents with too many permissions. Those risks are real, and I covered one of them in my piece on why AI agents going rogue is a governance problem. But the oldest category of risk didn’t disappear because the new one arrived.
The AI footprint is now part of the evidence
One detail stands out to me. Prosecutors say the planning left a trail in a chatbot account, and the defenders then used AI to analyse the device itself.
That cuts both ways for a business. AI tools used on company accounts can create a record that helps an investigation, and the same tools help security teams work faster on forensics. It also means you need a deliberate position on what’s logged, how long it’s kept and who is allowed to read it.
I’m not a lawyer, and the rules differ by country. In the UK, for example, staff monitoring needs to be proportionate and transparent, so take advice before you switch anything on.
Where I see founders over and under invest
| Control area | What I usually see | Question I’d ask |
|---|---|---|
| Model and data security | Well covered, board attention is high | Which data can reach which model? |
| Agent permissions | Patchy, often owned by one engineer | Who can stop each agent mid-task? |
| Access reviews | Done once, at set-up | When did someone last review who can see what? |
| Joiners, movers and leavers | Leavers handled, movers forgotten | What happens to access when someone changes role? |
| Physical security | Assumed, rarely tested | When did anyone last check meeting rooms and shared devices? |
| Logging and review | Logs exist, nobody reads them | Who looks at the logs, and how often? |
None of this is glamorous. That’s exactly why it gets skipped.
Five questions for your next board meeting
If you sit on a board or run one, I’d put these on the agenda:
- Who has access to the company’s most sensitive information, and when was that last reviewed?
- What happens to someone’s access when they change role or hand in notice?
- Is AI tool use on company accounts logged, who reads it, and have staff been told?
- When did anyone last check meeting rooms and shared devices?
- Could the company spot a trusted insider behaving oddly, or only an outsider?
If you can’t answer three of the five quickly, that’s your starting point.
Treat insider risk as a control, not an insult
The reason founders avoid this topic is that it feels like distrust. It shouldn’t.
Good access controls protect honest staff as much as the company. They mean nobody is blamed for something they couldn’t have touched, and they make an investigation faster and fairer if something does go wrong.
The AI era has added plenty of new things to worry about. It’s still worth remembering that the most reliable way into a company has always been through someone already inside it.
If you want a second opinion on how your risk reporting balances AI threats against people and access controls, that’s the kind of work I do under VAQA’s Operational Efficiency and Board Advisory pillars. Get in touch if you’d like to talk it through, or read more about what I advise on.
Frequently Asked Questions
Has the engineer been convicted?
No. Prosecutors have charged the engineer, but these are allegations and the case hasn’t been decided. Reports also say it hasn’t been tied to any foreign government.
What is insider risk?
Insider risk is the chance that someone with legitimate access, such as an employee or contractor, causes harm to the company, whether deliberately or by accident. It sits alongside external threats rather than replacing them.
Is it legal to monitor employees’ AI usage?
It depends on where you operate and how you do it. In the UK, monitoring generally needs to be proportionate, transparent and justified, so speak to a lawyer and tell staff clearly before you start.
Do smaller companies need physical bug sweeps?
Probably not as a routine. What they do need is basic hygiene: control who can enter sensitive rooms, notice unfamiliar devices, and review access when people move or leave.
Does VAQA advise on this kind of risk reporting?
Yes. Board risk reporting sits under Board Advisory, and the access and process controls sit under Operational Efficiency, both of which I work on through VAQA.
Tom Batting is a Forbes 30 Under 30 entrepreneur, founder of Obby and Baluu, and founder of FirstMotion. He advises founders and leadership teams through VAQA on board advisory, growth, go-to-market, AI implementation, operational efficiency, and finance and fundraising.