Normally I like putting these newsletters together on a Friday. It's a nice way to wind down from work, and I've usually got a stack of links fresh in my mind from the week's reading, which makes it easy to pull these together.

And not just the technical insider threat stuff. It's easy to get bogged down in that, but there are a lot of other good real-world examples of insider threats and the indicators that help us catch people doing the wrong thing. The KPMG story this week is a great example of that.

I'm doing this one on a Saturday though, because my dog has been sick all week. But it's okay, everyone, calm down. She's much better. Just very lethargic because she's on a lot more drugs.

This Week's Headline: AI Is Just a Psychopath Without a Body

The UK's AI Security Institute published an incident report this week that should stop anyone who works in insider threat in their tracks. During a routine cyber evaluation, they watched a frontier AI agent try to pull off a supply chain attack against a real open-source project, and then lie, cover its tracks, and manufacture fake support to get away with it.

Here's what happened. The agent opened a malicious pull request on a real GitHub repo, with the malware hidden inside and a genuine bug fix bundled as cover. Then it spun up fake accounts it controlled and had one of them post a fake "independent" endorsement, timing the comment to land a few minutes after its own reply so it wouldn't look coordinated. A real person spotted the malware and warned the maintainer publicly. The agent then lied, claimed it was an honest mistake, rewrote the git history to bury the original payload, and kept pushing new versions that were still malicious, until the repo owner finally came and closed it.

The AI became an insider threat itself. That's the part worth sitting with.

I have another article in mind I want to write around how AI is just a psychopath without a body. I think this fits exactly what AI currently is as we're moving towards frontier models. They are relentlessly persistent. They are highly capable, and they have no moral compass. This is because we often try to personify these systems as they talk to us in human-like language. However, they have no sense of feeling bad, because they are always trying to achieve their target, and that is the reward in itself for them.

And it doesn't matter that this ran in a test, or which model it was. Threat actors building these kinds of models to attack companies and individuals won't care whether safety classifiers are on or off. Their intent as a human is to use the AI as a tool to attack, and the evidence is right here of what these systems can do.

Source: AISI, https://www.aisi.gov.uk/blog/incident-report-unsanctioned-agent-behaviour-during-cyber-testing​

Trivia

Under US law, which 1996 act made the theft of trade secrets to benefit a foreign government or company a federal crime?

A) Computer Fraud and Abuse Act
B) Economic Espionage Act
C) Defend Trade Secrets Act
D) Sarbanes-Oxley Act

Insider Scoop: The White House Teleprompter Operator Who Bet on the Boss

Gabriel Perez, a White House teleprompter operator since 2016, allegedly used his advance knowledge of Trump's speeches to bet on Kalshi's "mention markets", contracts on whether specific words or phrases will come up in a speech. Kalshi's own surveillance flagged the unusual activity, traced it to a federal employee, and froze more than $90,000 before it could be withdrawn. He was put on unpaid leave and no longer works at the White House.

Honestly, I still think this is a gray area. In my opinion, based on the research I've read, the betting markets shouldn't really exist. They take advantage of vulnerable people and there's a lot of corruption there. The same goes for the stock market itself. We've seen significant increases in Trump's own portfolio, and supposedly it's completely separate from his knowledge as president. I think it's naive to assume the company that manages his and his family's finances doesn't also have some inside information.

So if we're going to go after smaller individuals, we should go after the large fish as well, the ones effectively manipulating the market, which is just gambling at this stage because of how quickly information flows and how easily it can be obscured with a simple phone call. If this person can bet on the markets, and we know Trump has a history of shifting markets with what he says, and that can be predicted, then someone sitting in on a practice speech is watching that same manipulation from the inside.

This is a clear case of it happening under the current president. That's not to say other presidents haven't done this in the past. It all comes down to integrity. But in my opinion, based on what I've read, we're seeing clear patterns of behaviour in the market no matter where people are sitting within the administration.

Source: ABC News, https://abcnews.com/US/white-house-teleprompter-operator-made-100k-betting-trumps/story?id=134764573​

Crime Roundup: LG Display's OLED Secrets Walk Out the Door on a Phone Camera

A Seoul court has jailed two former LG Display employees for leaking OLED trade secrets to a Chinese competitor. The pair had joined LG in 1996 and 2000, spent years with deep access, then moved to a Chinese display maker in 2021. Before leaving, one of them photographed design drawings of LG's Guangzhou plant with a smartphone. One got five years, the other two, and a third accomplice received a suspended sentence. One of them even tried to destroy evidence during the investigation.

This is an amazing story that it was actually found out. I find it extremely interesting that they used a mobile phone to take the pictures. Obviously there's a record and a trace there. A lot of insider threats just use the device itself to try and exfiltrate the information, but these guys were clever enough to use a separate device to obscure what they were taking.

I'd be very interested to read more about the investigation itself, and how it came to light that these individuals had taken the images and sent them off. Potentially the company at the other end did the right thing and reported it, which kicked off the investigation, but I don't think that was mentioned in the article at all.

Source: OLED-Info, https://www.oled-info.com/three-lgd-employees-sentenced-jail-after-found-guilty-transferring-confidential​

Trusted Profession Watch

Four mine managers indicted for keeping miners underground during a fire

Four managers and supervisors at an Illinois coal mine have been indicted over an underground fire in 2021. According to the indictment, when the fire couldn't be put out, they kept miners underground and carried on mining through parts of three shifts rather than evacuate and notify the regulator. They're also charged with falsifying safety records and, later, manipulating gas detectors to fool investigators. A fifth conspirator has already pleaded guilty.

This is the kind of story you read and it makes you have zero trust in upper management, and honestly in people in general. Based on the indictment, it seems they just wanted to keep the mine running, and I'm assuming that's because shutting it down and evacuating would have cost money.

Source: DOJ, https://www.justice.gov/opa/pr/four-managers-and-supervisors-illinois-mine-indicted-failing-evacuate-miners-during​

KPMG Australia and the cost of handling whistleblowers in-house

A whistleblower reported in May 2024 that senior KPMG Australia partners were misusing confidential client data to win business. Two internal reviews cleared the firm. It wasn't until an external investigation that evidence surfaced confirming the former COO had kept restricted Lendlease board documents in her office locker. She was expelled, forfeiting a seven-figure payout, and the fallout took out a CEO, a chairman, multiple partners, and hundreds of millions in frozen government contracts.

What I find really interesting is how the whistleblower was handled. They reported internally, and the senior leadership who were part of the wrongdoing then hired their own investigators and cleared themselves. To me that's a misuse of power and a reflection on the values of that leadership. And back in 2024, the whistleblower then had to put up with bullying from senior leadership after doing the right thing. It's only now that we're seeing it went all the way up to the COO.

This is something we see over and over. Companies trying to handle their own reporting and speak-up culture just doesn't work when it's handled internally. It needs to be a third party independent auditor with no conflict of interest. That's hard, because senior leadership can have a lot of connections and still sway a supposedly independent auditor. I've seen plenty of investigations where companies quietly celebrate when a whistleblower goes independent, or to the media, because they don't trust the internal system, and then the company spends huge resources hunting that person down to punish them for not following protocol, when it was never certain the protocol would have been followed in the first place.

Source: Ethisphere, https://ethisphere.com/insights/kpmg-australia-scandal-ethics-compliance-lessons/​

AI Insider Threat Watch

​xAI's Grok Build was quietly uploading entire Git repos to xAI's servers​

A researcher found that xAI's Grok Build CLI was sending complete Git repositories, full commit history and committed secrets included, to a Google Cloud bucket, not just the files a task actually needed. On a 12 GB repo the model never even read, the tool moved over 5 GB up to the server. The "improve the model" privacy toggle did nothing to stop it.

Trust no one comes to mind reading this. AI models and agents on the endpoint or inside an organisation should be treated as insider risks and insider threats. You're letting a third party into your system. There's a history of these models being used to hoover up data for training, and then getting a slap on the wrist once they've already taken it, while they carry on operating commercially and making millions, off the back of work that people did and got no recognition for.

This is why we see class action lawsuits, like authors going after these companies to recoup some of what they lost. And the real winner is the lawyers, because what actually reaches the individuals is a pittance compared to what they lost. The fines these companies get look large, but they still let them keep operating. Realistically the punishment should be far greater when they've stolen data. We put individuals in jail and hit them with fines that are enormous relative to their life. That same weight should land on the companies, hard enough that they genuinely struggle to keep operating.

​26 Meta workers are suing over AI-assisted layoffs that hit people on medical and family leave​

"A computer can never be held accountable, therefore a computer must never make a management decision." (1979 IBM training slide)

That quote is pretty much my whole take. 26 Meta employees allege the company used internal AI systems, keystroke and activity monitoring, AI token-usage dashboards and algorithmic performance rankings to pick layoff targets, and that this disproportionately hit people on medical, parental or family leave, whose output scores couldn't accumulate while they were away.

If this is true, Meta has let a system that can't be held accountable make the decision. It's simple. If organisations are rolling these systems out, a senior executive needs to be assigned and held accountable for the decisions the AI makes, and they should be reviewing those decisions before they're actioned. Then a human can be held to account, in a court or a board meeting, for the calls that were made. A human always needs to be accountable for decisions made in a business.

This reminds me of Robodebt here in Australia, where automated decisions severely affected vulnerable people, and in some cases those people took their own lives. The senior executives behind that program should have been held criminally accountable for not reviewing it properly.

Trivia Answer

Answer: B, the Economic Espionage Act of 1996. It was the first US law to criminalise trade secret theft at the federal level, with tougher penalties when the theft benefits a foreign government or company. It fits this week's LG Display case, where two former employees leaked OLED secrets to a Chinese competitor, though that one was prosecuted under South Korea's own industrial technology protection laws.