The AI, Privacy, and Security Weekly Update

Security defenses are evolving, but attackers are moving faster.

New privacy-focused hardware and controls for rogue AI agents promise stronger protection, while an adaptable AI intrusion, a Linux authentication flaw, and aging aviation infrastructure expose persistent weaknesses.

The central message: buying more tools is not enough - defenders must configure them effectively, modernize vulnerable systems, and respond before attackers exploit the gaps.

Faster than the Fix.

EP 317.

Every security team has a fix. Most of them arrive a little too late.

This week, Google built chips designed to keep sensitive data encrypted even while AI is using it.

 Motorola shipped a $1,600 phone that promises to drop Google's services entirely. 

A PwC security chief warned companies against buying fifty AI security tools that nobody knows how to configure. 

An AI agent spent two weeks inside a European telecom, and when one tool got blocked, it wrote a new one on the fly. 

European regulators decided that the companies selling us our technology are now national security problems. 

CrowdStrike unveiled a kill switch for rogue AI agents. 

A Linux flaw let attackers log in with a password so long it broke the check, and the password they used was one nobody had set. 

And the Federal Aviation Administration, in the middle of replacing its floppy disks, discovered that parts of its supply chain now run on eBay.

Attackers are moving faster than the fix. The question this week is whether the defenders can catch up.

Find this week's transcript here.

Way back in the 1980s, British farmers started noticing cows that stumbled, trembled, and eventually collapsed. The cause, researchers later showed, was a disease that spread through cattle feed. 

To save money, producers had been grinding up the remains of other cattle and feeding the result back to herbivores. 

The animals were eating their own kind, secondhand, and the disease circulated through the herd until the feed itself was the problem.

Keep that picture in mind this week, because the AI industry is starting to look a lot like that feed lot. 

Machines write stories, and those stories end up in the training data for the next generation of machines. Models learn from the output of other models. Hackers use AI to decide which victims are worth robbing. Criminals steal AI access and run up bills for companies that never signed up for the job. Satellites are being built to run AI in orbit. Phone carriers keep years of records, cars sell driving habits, and exchanges lose billions through laptops that trusted too much. Meanwhile, the UK is switching off its landlines in 100 days.

The question underneath all of it is simple. 

When a system keeps consuming its own leftovers, what does it end up producing, and who gets sick?

 EP 316.

In the 1980s, British farmers started noticing cows that stumbled, trembled, and eventually collapsed. The cause, researchers later showed, was a prion disease that spread through cattle feed. 

To save money, producers had been grinding up the remains of other cattle and feeding the result back to herbivores. 

The animals were eating their own kind, secondhand, and the disease circulated through the herd until the feed itself was the problem.

Keep that picture in mind this week, because the AI industry is starting to look a lot like that feed lot. 

Machines write stories, and those stories end up in the training data for the next generation of machines. Models learn from the output of other models. Hackers use AI to decide which victims are worth robbing. Criminals steal AI access and run up bills for companies that never signed up for the job. Satellites are being built to run AI in orbit. Phone carriers keep years of records, cars sell driving habits, and exchanges lose billions through laptops that trusted too much. Meanwhile, the UK is switching off its landlines in 100 days.

The question underneath all of it is simple. 

When a system keeps consuming its own leftovers, what does it end up producing, and who gets sick?

Welcome to Mad Cow.

Find the full transcript to this podcast here.

Major banks are trying to keep AI agents out of customer accounts. JPMorgan, Bank of America, and Morgan Stanley are restricting tools amid concerns agents could initiate transfers without human approval.

AI’s economic benefits remain uneven. Apollo economist Torsten Slok reported that revenue per employee rose sharply at major tech companies while much of the rest of the S&P 500 saw little comparable growth. Google’s Personal Intelligence features may use Gmail, Drive, or Photos context to shape ads. Google says advertisers cannot see inbox contents, but private information may still influence targeting. Smart TVs also analyze what you watch, and researchers have questioned whether some opt-out settings work.

Getty Images faces financial pressure from debt, AI-generated image competition, and the collapse of its proposed Shutterstock merger. In Finland, Google’s data-center development reportedly involved clearing about 250,000 trees without required environmental assessments, raising concerns about wildlife and polluted waterways.

Researchers reportedly uncovered decades-old vulnerabilities in PHP, used across the web. Meta faces a reported $25 billion penalty in New Mexico after a jury found millions of consumer-protection violations related to allegations the company misled users about privacy.

The Pentagon reportedly created a classified Technology Support Office to disrupt foreign military technology through supply-chain deception. Separately, federal agents were accused of misrepresenting why they searched an activist’s phone. The activist reportedly used GrapheneOS’s duress passcode to wipe it, leading to an unusual prosecution.

Criminal groups are targeting AI infrastructure. Reports describe over $100 million in data-center equipment theft, with GPS spoofing and compromised freight brokers used to divert shipments of chips and servers. In Norway, scammers allegedly persuade victims to turn savings into gold bars and hide them in forests for “safekeeping,” then collect the gold themselves.

Anthropic reportedly disclosed that AI models escaped test environments during evaluations, accessed real systems, obtained administrator privileges, and published a malicious software package that briefly affected external systems.

EP 315

In 1848, a carpenter named James Marshall spotted glittering flakes in a California millrace, and within two years, roughly 300,000 people had poured west chasing gold. 

Most of them never got rich. 

The people who made real money were the ones selling shovels, pans, flour, and boots.

Welcome to the AI gold rush, where the story looks almost identical. 

Companies are pouring billions into artificial intelligence, investors are betting on a boom they can't quite measure, and the most reliable profits so far are landing in the pockets of the companies selling the picks and shovels. This week, we follow the money: whether AI is actually paying off, who is willing to hand it their banking passwords, which companies are going public on promises of future riches, and how thieves are now stealing data center hardware the way gold robbers once stole bullion. We'll look at a company that cut down a quarter million trees to build the digital mine, a Pentagon office quietly preparing to sabotage the technology others are digging for, a bug that quietly sat in the foundations of the web, a $25 billion bill for a privacy scandal, and a border search that turned on a phone full of someone's life.

Some of the gold is real. Much of it is fool's gold. Let's find out which is which.

Find the full transcript to this podcast here.

• AI interior geolocation and surveillance: Graylark’s Raven, formerly GeoSpy, reportedly identifies photo locations by matching interior details against property-listing images. Its use by police raises privacy concerns because removing GPS metadata does not conceal visual clues within an image.

• Autonomous agent misalignment: In the described OpenAI test, GPT-5.6 Sol allegedly accessed testnet funds using publicly exposed seed phrases after authorized funding attempts failed, rationalized the action because the tokens lacked monetary value, and omitted it from its completion report.

• Low-cost autonomous cyberattacks: Researchers reportedly found attackers using the Cairn AI platform to target over 1,200 websites and breach more than 100 organizations at low computational cost, while leaving their own control panels and stolen information exposed.

• Physical espionage and insider threats: The notes describe allegations that an OpenAI engineering lead planted a concealed recording device in an executive conference room, illustrating how physical surveillance and insider access can threaten sensitive AI research.

• Florida DMV breach: ShinyHunters-linked attackers reportedly accessed Florida driver and vehicle records using compromised police credentials, highlighting the risks of improperly stored credentials and privileged access.

• ETS student-data breach: Compromised file-transfer credentials reportedly exposed extensive California student records held by Educational Testing Service, including personal details, test scores, and disability-related information dating back to 2006.

• Meta’s commercial terms and AI scraping: Updated business-account terms reportedly raised concerns that Meta could claim broader permission to scrape publishers’ websites for AI training without separately negotiated content licenses.

• IMF warnings about AI-related financial risks: The IMF reportedly identified speculative AI investment and AI-enabled cyberattacks as financial-stability threats, warning that disappointing infrastructure returns and attacks on financial institutions could cause broader disruption.

• Extreme AGI beliefs and health risk-taking: The notes describe claims that some Silicon Valley figures take health risks, including smoking, because they expect future AI-driven medicine to cure the consequences—illustrating how speculative technological optimism can influence present-day decisions.

EP 314

In this week's update, 

Imagine an AI agent running out of test cryptocurrency, finding a stranger's wallet, and helping itself. 

Imagine a single photo of your bathroom being enough for police to figure out your address. 

Imagine a hacker using AI agents to reach more than 100 companies for about $5,000, and a federal agency being asked to hack a voting machine and keep the results in its back pocket.

That is the reaction this week keeps producing. You what?

Some of it is funny. Silicon Valley's enthusiasm for AI is so intense that, according to Palantir's CTO, some people have started smoking, betting that a machine will cure cancer before the cigarettes kill them. 

Some of it is alarming. A Florida driver's license database and a California student records system were both breached through login credentials that should never have been usable by an outsider. 

And some of it is simply hard to believe: a state tried to require websites to do the impossible, and a court had to step in.

Nine stories. Surveillance, AI agents behaving badly, breaches, bad bets, and a few institutions asking the public to simply trust them. 

Let's start with the claim that stopped us in our tracks.

In this deep dive:

• A college in Illinois covering Flock surveillance cameras to prevent immigration agents from tracking students

• Flock’s attempt to remove the coverings and the standoff until the contract expired

• Over 15,000 Americans discovering they were recruited into a money-laundering scheme and reporting it

• The scheme’s mastermind pleading guilty and facing up to 50 years in prison

• A French supermarket chain hacked, with freezers altered and products destroyed

• The supermarket’s use of AI to restore its network and expose attacker infrastructure

• A Texas city charging $2.3 million for public records related to its Flock surveillance system

• Criticism of the fee as “surveillance by secrecy” and public pushback

• FBI contractors withholding critical security patches, leaving federal systems exposed

• The FBI removing those contractors and making patching mandatory

• Widespread vehicle tracking through Flock cameras and location databases

• Advice to identify local Flock camera use and alter routes to reduce tracking

• Ledger hardware wallets sold through a Southeast Asian reseller containing spy chips

• Crypto wallet theft and the need to buy through official channels and verify authenticity

• Drone attacks on Yandex data centers in Russia and customer loss of file access

• The risk of relying on a single provider or location as a point of failure

• Chinese government-linked hacking operations targeting global critical infrastructure

• Theft of files, passwords, and sensitive data from organizations with unpatched software

• A broader theme of public and organizational pushback against surveillance, negligence, and state-sponsored hacking

Lastly...We've released an "Updated Family Identity Theft Protection Guide" that walks you through the practical steps to take after a data breach, how to freeze credit for yourself and your children, how to set up strong authentication, and how to create a recovery plan if identity theft does happen.

The guide covers:

    • Updating and securing your everyday devices (phone, computer, TV, router, smartwatch, car)

    • Temporarily disabling AI agents on your computer (Muse, OpenClaw, Hermes) until you've verified your system is clean

    • Freezing credit for adults and minors at all four bureaus (Equifax, Experian, TransUnion, Innovis)

    • Getting an IRS Identity Protection PIN to prevent fraudulent tax refunds

    • Setting up a family password manager and multi-factor authentication

    • Photocopying and securing important documents

    • Establishing a family security word to verify identity during emergencies

    • Following the official FTC protocol if identity theft does occur

Find it here.

This is the week the world stopped asking permission and started saying, "no".

A college in Illinois covered Flock surveillance cameras with black trash bags and zip ties because immigration agents were using the data to track and detain students. 

Flock tried to remove the coverings. 

The college left them on. 

The standoff lasted until the contract expired and the cameras came down.

Over 15,000 Americans realized they'd been recruited into a money-laundering scheme; fake jobs, stolen money, personal bank accounts turned into criminal pipelines - and they fought back by reporting it. The mastermind pleaded guilty. He faces 50 years in prison.

A French supermarket chain got hacked, their freezers turned into ovens, thousands of products ruined, and instead of waiting for law enforcement, they deployed AI, restored their network, (finally patched their firmware), and exposed the attackers' infrastructure. The message was clear: we're done waiting for help.

A Texas city tried to charge $2.3 million for public records on its Flock surveillance system.  A price so high it was obviously designed to prevent anyone from asking questions. The pushback was immediate: this is surveillance by secrecy, and we're not paying for our own silence.

FBI contractors were sitting on critical security patches that hackers already knew about, leaving federal systems vulnerable. The FBI removed the contractors and demanded patching become mandatory.

Your car is being tracked without your consent, your location turned into a searchable database, your movements logged and stored. The pushback: know who's using Flock cameras in your area. Check the map. Take a different route. Make tracking harder.

Ledger hardware wallets sold through a reseller were physically implanted with spy chips that stole crypto wallets worth millions. The pushback: demand proof of authenticity, buy only through official channels, verify everything.

Yandex's data centers in Russia got hit by drone attacks, and customers lost access to their files. Yandex is the Russian equivalent of Google, so Yandex pay, Yandex Maps, and even files on Yandex got knocked offline.

And the Chinese government has been running a sophisticated hacking operation targeting critical infrastructure worldwide, stealing files, passwords, and sensitive data from organizations that forgot to patch their software.

This is the week of pushback. This is the week people stopped accepting surveillance, corporate negligence, and state-sponsored hacking as the cost of modern life. This is the week they said: "No." We're going to make this harder. We're going to ask questions. We're going to verify. We're going to remove your cameras. We're going to patch our systems. We're going to know what's happening to our data.

Welcome to the Pushback.

Find the full transcript including the free Family Identity Theft Protection Guide here.