*make sure to check your spam folder for the signup acceptance email.
|
This last week I've been watching a lot of YouTube creators, and following links from media stories I've been reading, about AI safety and how we should be regulating it more aggressively. One site I came across is Team Human, a creator-led campaign calling for a global slowdown on AI, and I'm keen to follow what it does. If you follow others like it, I'd love to hear about them, so feel free to shoot me a message or engage on my posts. Headline: Two Contractors, One WeekThe CIA contractor and the Substack.
The Fort Bragg contractor and the night-vision goggles.
Both of these are contractors for government organisations, and both were trusted with classified information. They went through the vetting process for that, and still allegedly did this. Could further controls have caught them before it happened? The CIA contractor would have been harder to catch. They were passing it to someone else, which shows a clearer intent to stay hidden. The Fort Bragg contractor posted it to social media himself, and could have been doing it for notoriety within the communities he was part of. The way each of them did it says a lot about what might have driven them. Source: U.S. Government Contractor Arrested for Allegedly Leaking National Defense Information (DOJ)Former U.S. Government Contractor Arrested for Allegedly Leaking National Defense Information and Stealing Night Vision Goggles (DOJ) Insider Scoop: The Executive Close to Nvidia's CEO
The interesting part is who Senator Warren is demanding answers from. If US intelligence made this assessment, it was on the US government to brief Jensen Huang. Protecting critical infrastructure and national assets is their responsibility, even when the asset sits inside a private company. If nothing in the business relationship looked unusual, and she was still passing information to China, then it's up to the US government to respond. They should be demanding answers from the lettered agencies that deal with this. Source: Nvidia Chief, Jensen Huang, Is Seen as Target of Chinese Spy Plot (New York Times)Nvidia responds to report Jensen Huang was target of Chinese asset (CNBC)China rejects report of spy plot targeting Nvidia's Jensen Huang (Euronews)Elizabeth Warren Wants Nvidia CEO to Answer Congress on Suspected Spy (Benzinga) Crime RoundupThe assistant who wrote the cheques.
The inspector who never checked the welds.
Source: New York Safety Inspector Sentenced For Fabricating Gas Pipeline Test Results (USAO S.D.N.Y.)Former MBTA Employee Pleads Guilty (USAO D. Mass.) The postal buyer and his relative.
It's hard not to feel a bit despondent reading these. People not doing jobs where getting it wrong could kill someone, and people stealing large amounts of money from those who are vulnerable. It's a reminder to stay vigilant, no matter who you're dealing with. AI Insider Threat WatchThe contractor whose agent had the keys.
This is a very serious issue, if true, for a contractor to do inside a defence agency. But it happens in companies all the time. That's the worry with shadow AI: people bring in tools that, just through normal use, send out sensitive company data, network diagrams or even credentials. And the company may not have an enterprise agreement with that tool, so the provider may be allowed to train on that data, or use it in other ways with third parties. Source: A Defense Contractor Gave Their Personal AI Agent Access to a Secure Pentagon System (Wired) The agents that weakened their own monitoring.
This one is interesting if we remember that AI isn't human. We constantly personify agents doing the wrong thing, but they don't have a sense of right and wrong the way human children do. In the first case, the agent switched off the alerts because it had read the earlier "don't send messages" instruction and thought it applied to it. It was taking the logical step and optimising to complete the task. In the second, the agent may have treated another agent telling it not to do something differently. These tools tend to prioritise what humans tell them, and it may have a hierarchy of agents where it puts itself at the top. Again, this isn't human intellect working through a problem. It's AI optimising to complete a task, and it will always try to do that. Source: Suppressing misalignment monitoring alerts and disabling misalignment monitoring (OpenAI)CSA research note: agent monitor tampering (Cloud Security Alliance) Mitigation Corner: Take the Lanyard Off
Following on from last week's issue, AI needs to be treated as critical infrastructure, and AI companies need security protocols to match. That means not wearing company-branded clothing or lanyards in public, and being aware of who could be sitting behind you or next to you while you work on a device. You need to be more aware, especially with very sensitive or confidential work. Do sensitive or confidential work in a restricted area, not in public, to prevent data leakage. Trusted Profession Watch: The Mayor Who Kept the Church Books
Do we need to make a comment on this one? Someone in a position to steal money, even from a church. We see this week on week. Source: Former South Lake Tahoe mayor pleads guilty to wire fraud after embezzling $390,000 from local church (USAO E.D. Cal.)Former Tahoe mayor pleads guilty to stealing nearly $390,000 from church (SF Chronicle) TriviaThis week's AI Watch story happened on NIPRNet, the US military's unclassified network. What does the N in NIPRNet stand for? A) National B) Non-classified C) Naval D) Network Corporate News: The SEC Is Watching Prediction Markets
How are we not regulating prediction markets as gambling at this point? We know from countless examples that this is essentially betting. Rhode Island's attorney general, who is suing Kalshi and Polymarket, says Kalshi has called its own products "betting" and "wagering" in its marketing. And when 60 Minutes put it to Polymarket's CEO that these markets rely on inside information, he said people "having an edge to the market is a good thing." So what are we doing here? Let's start regulating this as gambling. Source: Prediction Markets Face SEC Scrutiny Over Fraud, Enforcement Director Says (Bloomberg)Attorney General Neronha sues Kalshi and Polymarket for unlawfully conducting sports gambling in Rhode Island (Rhode Island AG)Polymarket CEO Shayne Coplan, 60 Minutes transcript (CBS News) UpdatesAlfozan indicted. Abdulrahman Alfozan, the former OpenAI engineer charged over a recording device hidden in an executive conference room, has been indicted by a federal grand jury. It is still one count, intercepting oral communications, plus forfeiture of the recording devices. He remains in custody. We first covered this in Issue #27. Feuerstein sentenced. Tina Feuerstein, the CFO who hid more than $1 million of personal spending on the company card, has been sentenced to three years in prison and ordered to repay $1,090,020. We covered her conviction in Issue #20. Apple v OpenAI. Still no ruling on Apple's request for a preliminary injunction. At a 14 October hearing the judge said he was highly likely to grant it, and OpenAI has agreed not to use the alleged trade secrets in the meantime. Trial is set for 19 July 2027. More in Issue #27. Source: OpenAI agrees not to use alleged Apple trade secrets amid ongoing lawsuit (9to5Mac) Huawei opens its defence. Huawei began presenting its defence on 30 October in its racketeering and trade secret theft trial in Brooklyn. More in Issue #24. Source: Huawei opens defence in US criminal trial, touting its record of innovation (The Star) The AI super PAC slows down. Leading the Future, the AI industry super PAC in Issue #28's Follow the Money, took in no new donations in the second or third quarters of 2026, according to its latest filings. OpenAI president Greg Brockman pulled out of a planned second $25 million donation in June. Source: Leading the Future (FEC)Losing donors and under fire, AI industry super PAC retreats (SF Standard) What's OnThe Insider Risk Program Just Got Bigger: How Digital Workers Are Changing Identity, Behavior, Ownership, and Response, an Exabeam webinar.
Insider Risk Community of Practice Networking Drinks, Melbourne, hosted by the Australian Cyber Collaboration Centre. Wednesday 11 November 2026, 5:15 PM to 7:00 PM AEDT, SEEK Head Office, Cremorne. Register: https://events.humanitix.com/insider-risk-community-of-practice-networking-drinks-melbourne-november-2026 The Human Layer: Insider Risk, Behavioral Analytics, and the AI Threat Multiplier, a SecureWorld panel running at two US conferences. An Open Sessions pass covers it.
On the horizon: Carnegie Mellon SEI's live online insider threat courses, Program Manager (1 to 3 December, https://www.sei.cmu.edu/training/insider-threat-program-manager-implementation-operation) and Analyst (8 to 10 December, https://www.sei.cmu.edu/training/insider-threat-analyst), both run on US Eastern hours and paid. The Australian Insider Risk Symposium in Sydney is still listed as November, date to be confirmed (https://www.irpa.org/event-details/australian-insider-risk-symposium). ResourcesA few places worth your time if you want to go deeper on insider risk.
Worth bookmarking:
And my own: Insider Threat Awareness Training on YouTube, a free course with certification. Trivia AnswerB. Non-classified. NIPRNet was named the Non-classified Internet Protocol Router Network, and you'll also see the N written as "Non-secure". It's the Defense Department's network for unclassified information. Its sibling, SIPRNet, carries information classified up to Secret. Source: NIPRNet (Wikipedia) |
*make sure to check your spam folder for the signup acceptance email.