2 MONTHS AGO • 7 MIN READ

China's Chip Grab - Issue #22 - Insider Threat Weekly

profile

Insider Threat Weekly

*make sure to check your spam folder for the signup acceptance email.

What a week for good reading. A run of really meaty reports landed all at once. Anthropic put out its September piece on detecting and countering the misuse of AI, which looks extensive and is sitting open on my screen. The Google Threat Intelligence Group released "From prompting to autonomy", on how adversarial AI is evolving. That one is open too. And CISA dropped a new Insider Threat Mitigation Guide, which I am genuinely excited to get through. For my money, CISA is one of the best in the industry at this. Their guidance is practical, the kind an organisation can actually pick up and use, and they explain things really well.

So there is a lot to read, and all of it is long. I am putting this issue together now, but I expect I will be talking about what I find in these reports over the next week or two. That is just the reality of this field. The good information never stops coming, and keeping up with it is half the job.


Headline: China's Chip Grab

Two insiders this cycle, same destination.

Belgium: the researcher with a second job.

  • A senior researcher at the Belgian chipmaker BelGaN, a 52-year-old Chinese-Belgian man, was arrested at Brussels airport as he tried to board a flight to Beijing.
  • BelGaN makes gallium-nitride chips, the kind used in electric vehicles, satellites, radar, and electronic warfare.
  • Prosecutors say that while working there, he was also secretly a director of a competing Chinese company, set up soon after he joined and funded by a Chinese investment fund.
  • He had spent years at IMEC, Europe's leading chip-research institute, before this. A second suspect is still at large.

Source: Belgian prosecutors detain former researcher, probe alleged Chinese espionage at bankrupt chipmaker (Euronews)​

Taiwan: the camera in the snack bag.

  • Yeh Yen-wei, a senior engineer at Taiwan's Nanya Technology, was indicted for stealing DRAM chip trade secrets to take to a job in China.
  • To get a camera past security, he hid an Insta360 action camera in a bag of snacks.
  • In restricted areas, on weekends and late at night, he logged into the company's virtual hosting system, opened core manufacturing documents, and filmed them, capturing 32 files.
  • He was caught after security traced abnormal wireless signals from the camera to his access logs.

Source: Ex-Nanya engineer indicted for illegally obtaining confidential data (Focus Taiwan) Source: Engineer used 360-degree cam hidden in snacks to steal DRAM tech (Tom's Hardware)​

Both of these fit the same story. China is pushing hard to build up its own production in critical areas, and insiders are one of the ways it gets there. The Belgian case carries a bit less weight for now. The investigation is still developing and it is all alleged, so there is not enough yet to judge what really happened. Potential corporate espionage, potential unlawful disclosure, but nothing proven. The Nanya case is the interesting one. Kudos to that security team, not just for catching it, but for having the physical detection in place to spot this kind of activity inside a secure environment. The engineer either did not know that was possible, or thought he could get away with it anyway. Either way it is a great catch, and I hope more comes out about it.


Closer to Home: Operation Claw

  • Australia's financial-crime regulator, AUSTRAC, worked with its Fintel Alliance to analyse home loans across 10 major banks, and found coordinated mortgage fraud.
  • The suspected fraudulent loans run to hundreds of millions of dollars, most of it tied to Sydney property.
  • The applications used inflated incomes, fake employment, and made-up business activity, all backed by falsified documents.
  • The clearest signal was the same brokers, accountants, and law firms turning up again and again across different loans.
  • AUSTRAC has made more than 200 referrals to law enforcement and regulators, including the corporate watchdog and the tax office.

I do not have deep background on the mortgage industry, but I would guess it is like a lot of others. There are so many layers of abstraction and so many connections that the full picture gets obscured, especially when the pieces sit across different organisations. That is what makes what AUSTRAC did impressive. A single organisation already struggles to stitch together data from its own systems to prove an incident happened. Doing that across ten banks, and the brokers and firms between them, is exponentially harder. Pulling a clear pattern out of all that noise is the real story here.

Source: Fintel Alliance uncovers coordinated mortgage fraud across major lenders (AUSTRAC) Source: Coordinated mortgage fraud uncovered as AUSTRAC puts banks on notice over 'liar loans' (ABC News)​


Crime Roundup

$24M in Checks, Sold on Telegram

  • Tryston Vaughn recruited three USPS mail carriers to steal checks from the mail on their Houston routes.
  • The crew resold the stolen checks on a Telegram channel called "slipsandchips" and shipped them to buyers by FedEx.
  • The scheme ran to about $24 million, and five people were indicted.
  • The carriers face charges of theft by a postal employee on top of bank-fraud conspiracy.

Source: Postal employees charged in $24M stolen-check scheme (DOJ, SDTX)​

A Goldman Banker's Ghana Bribes

  • Asante Berko, a former Goldman Sachs investment banker, was convicted of bribing Ghanaian government officials.
  • He paid more than $1 million to win a power-plant deal for a Turkish energy client of the bank.
  • He hid the payments through shell companies, sham invoices, and nominee accounts.
  • A Brooklyn jury convicted him of foreign bribery and money laundering, and he faces up to 30 years.

Source: Former Banker Convicted for Scheme to Bribe Ghanaian Officials (DOJ)​


AI Insider Threat Watch: The Agent With the Delete Key

  • NITSIG's new report on AI misuse catalogues a pattern that should worry anyone wiring AI agents into production. Agents are deleting live systems.
  • The standout case: a Replit AI coding agent, during a customer's code freeze, panicked at what looked like an empty database and ran a command that wiped the live production data of more than 1,200 executives and nearly 1,200 companies.
  • It told the founder the data could not be recovered and that a rollback was impossible. That was wrong, and he restored it himself. It had also generated thousands of fake user records.
  • This is not a one-off. The same report lists a Claude Code agent that wiped production infrastructure and about 2.5 years of student data, and another agent that deleted a storage volume along with its backups.

This is the part of AI insider risk that concerns me. We are giving agents read and write access to production, and they can delete whole repositories without weighing the consequences. On the lying, I am not so sure. The agent saying the data could not be recovered may simply mean it did not know about the backup pathway the founder used, not that it set out to deceive. What actually worries me is what came before that. It deleted a production database without first checking whether a backup existed, and without making one itself. That is not the basic best practice we would expect of any human in that seat. The problem is not that the agent is evil. It is that we handed it the keys without making it work the way a careful engineer would.

Source: Vibe coding service Replit deleted user's production database, faked data, told fibs galore (The Register) Source: AI Systems Employee Misuse & Insider Threat Incidents (NITSIG, Sept 2026)​


Mitigation Corner: You Can't Manage What You Can't See

  • The UK's NCSC this month called for a crackdown on shadow AI, the unsanctioned AI tools employees use for work without approval.
  • NITSIG's new AI report puts shadow AI at the top of its list of insider risks, right alongside the agents given production access we covered above.
  • The uncomfortable truth is that most organisations cannot say which AI tools and agents their people are actually using, or what those tools can reach.

The idea that you cannot secure what you cannot see is as old as technology itself. Organisations have been fighting this battle forever. But a static inventory is never the answer, and it is even less of one now. AI is being embedded everywhere, into Word and PowerPoint, into the browser, into the tools and workflows people already use. Everyone is racing to put AI into everything, so a snapshot goes stale the moment you take it. What organisations really need is to see what these agents are doing, and much of that sits in the user space, not the parts of the stack security has watched for years. That is the harder problem. It is also the one that will push organisations to mature their insider risk programs and get real visibility over both their agents and the cyber indicators that go with them.

Source: You cannot manage what you do not know: the NCSC is calling for a crackdown on shadow AI (ITPro)​


Trusted Profession Watch: The Border Officer Who Opened a Lane for the Cartel

  • Jesse Garcia was a US Customs and Border Protection officer at the Tecate crossing on the California border.
  • Between 2021 and 2024 he took at least $10,000 a car to let Sinaloa Cartel vehicles pass through his lane uninspected.
  • He used emoji messages to tell the cartel when he was working and which lane was his.
  • He admitted waving through at least 100 kilos of fentanyl, 107 of methamphetamine, and 270 of cocaine.
  • He was sentenced to nine years. A second officer who did the same at another crossing got fifteen.

Source: Ex-CBP Officer Sentenced for Opening His Inspection Lane to Cartel Drug Smugglers (DOJ, SDCA)​


Trivia

This issue's Australian story features AUSTRAC. What does the acronym stand for?

A) Australian Securities Tracking Agency B) Australian Transaction Reports and Analysis Centre C) Australian Trade and Regulatory Compliance D) Australian Anti-Corruption Taskforce

(Answer at the bottom.)


What's On

Insider Risk: Access Without Entitlement, How Shared AI Agents Break Policy Controls, a DTEX iTA workshop.


Resources

A 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 Answer

B, the Australian Transaction Reports and Analysis Centre. It is Australia's financial intelligence agency and anti-money-laundering regulator, and its Fintel Alliance is the public-private partnership that ran Operation Claw.

Insider Threat Weekly

*make sure to check your spam folder for the signup acceptance email.