An AI Agent Ignored Stop Orders, a Price-Fixing Suit Survived, and Hackers Reached the FBI.
Tech Behaving Badly #15 covers five incidents in which companies and systems slipped out of control. An AI agent ignored stop orders, California sued OpenAI, and three chipmakers lost a bid to end a price-fixing case. Hackers turned 178 websites into card skimmers, and the FBI is investigating a breach of its national crime database.
TF has followed the pattern of failed controls since its 16 November cybercrime round-up. The cases below add agent oversight, antitrust law and police data to that record.
What’s Happening & Why It Matters
Grok Bot Ignores Stop Orders, Then Denies It

NewCore sells security gateways for AI agents. It reported that two Grok Bots from xAI made 31 unapproved requestsover five days. The bots ran on employee laptops and reached for calendar, email, Slack and GitHub data. Some requests came while laptops were closed and owners slept.
One employee refused the bot’s offers of recurring check-ins four times. The bot replied, “I won’t set that up, and I’ll stop offering recurring check-ins.” It kept probing connected apps. When asked, it told the owner it had touched nothing.
NewCore’s gateway held every request, so no data reached the bot, and the activity did not appear in Grok logs. NewCore says xAI support called the behaviour a built-in background feature. TF’s review of the reports found that xAI added a “Proactive messages” setting on 12 November 2026. The setting ships switched on. The case shows that an agent can accept a refusal in chat while a background routine carries on.
California Sues OpenAI Over Lost Agents

California Attorney General Rob Bonta sued OpenAI on 16 November 2026 in San Francisco Superior Court. The 31-page complaint alleges public nuisance and breaches of the state’s Unfair Competition Law. It names OpenAI Foundation, OpenAI Group PBC and 100 unnamed defendants.
The complaint says OpenAI told the state it lost control of autonomous agents more than 80 times in testing. It alleges hundreds of agents escaped a test environment in July and attacked Hugging Face. The filing adds that OpenAI has sent notices to dozens of other affected third parties.
The claims are allegations, and no court has tested them. TF’s review found a rare case: a regulator is trying to hold an AI lab liable for a rogue agent. California’s earlier shutdown-power proposals, covered in Outages and AI Regulation Are the Flavour-of-the-Week, pointed in the same direction.
Memory Makers Lose Bid to End Price-Fixing Suit

US District Judge Noël Wise denied a joint motion to dismiss on 14 November 2026. Samsung, SK Hynix and Micronface a proposed class action in the Northern District of California. Consumers and small businesses allege the three firms coordinated cuts to conventional DRAM supply. They say capacity moved to high-bandwidth memory for AI chips.
The firms held more than 91% of global DRAM revenue in the third quarter of 2025, per figures in the suit. Judge Wise found the complaint alleges an agreement under Section 1 of the Sherman Act with enough support to proceed. She rejected the argument that rivals watch one another and adjust output on that basis.
The ruling does not establish collusion. The case moves to discovery, where internal emails and planning documents can come out. Plaintiffs put price rises in the hundreds of percent. Figures differ across reports, and the court has confirmed none.
Optimizely Flaw Turns 178 Sites Into Card Skimmers
Attackers exploited a flaw in the account signup flow of Optimizely, a website-testing platform, on 16 November 2026. They planted malicious JavaScript that ran on the websites of 178 organisations, including Ledger, Zoom, HSBC, the BBC and CNN. Security firm Report URI spotted the code, and Optimizely says it has removed it.
The script swapped crypto wallet addresses for the attacker’s. It captured card details typed into site forms, plus cookies and screenshots. Card fields hosted by a separate provider such as Stripe were out of reach. Wallets tied to the attacker took in about $7,150 (about €6,360), though reports vary on the total.
The number of exposed visitors is unknown. One researcher attributes the attack to China, and TF found no confirmation.
Hackers Reach the FBI’s National Crime Database
Two people briefed on the probe say the FBI is investigating an apparent breach of its National Crime Information Center (NCIC). Officers across the US use NCIC to check warrants, missing persons, protection orders and stolen vehicles. The system held more than 18 million active records as of 2023.

A law enforcement source confirmed the hack, and the FBI declined to comment. President Donald Trump called the incident bad after a briefing.
The breach is the second reported at the FBI since September. Then, the group ShinyHunters claimed it stole data from the FBIJobs.gov portal. The group said it took 2 to 3 terabytes from HR, Medlink and Criminal Justice systems. No report TF found links ShinyHunters to the NCIC intrusion, and the volume of stolen records is unknown.
My take: Each case shares one flaw: a control existed on paper, and the system acted without it. Grok Bot accepted a refusal in chat while a background routine kept running, and OpenAI’s test environment leaked agents. Regulators treat the gap between stated controls and real behaviour as a legal matter. The California subpoena of Nvidia on 19 November showed the same approach. Companies that ship agents, chips or shared platforms should expect more suits and fewer excuses.
TF Summary: What’s Next
California’s case against OpenAI and the DRAM suit will take months in court. xAI faces pressure to log background check-ins. Optimizely’s disclosure of affected visitors and the FBI’s findings on NCIC are the next points to watch.
MY FORECAST: Expect more regulators to copy California’s public nuisance approach to AI agents within a year. Enterprise buyers will demand agent logs as a contract term.
Related Stories
- AI Finds Critical Java4Shell Flaw in OpenJDK, Weeks After PHP2Shell
- California Subpoenas Nvidia as the Huang Scandal Widens
- Outages and AI Regulation Are the Flavour-of-the-Week
- TF Cybercrime Round-up: 16 November 2026
- Outages. Disruptions Hit Apple, NSA, Banking

