404 Media: Leak Shows How Police Get Around Protections on Locked iPhones

The 404 Media podcast episode dated October 8, 2026 has three stories, all about how technology behaves in human hands. According to a leak, police are bypassing an important iPhone protection. Someone has built a "torture chamber" for language models. Lawyers have once again gone overboard with ChatGPT. The editors describe each story in a single line, with no technical detail, so below we separate what the announcement states directly from what is known background.

What we know about the iPhone hack

The episode's main claim: law enforcement has found a way to bypass a "very important" iPhone security feature, and a leak is the basis for that claim. The announcement does not say which feature it means. Naming one without the full text of the story would be guesswork.

The context, however, is known. iPhone protection has many layers: data encryption with a key tied to the passcode, USB Restricted Mode, which disables data transfer over the cable on a device that has not been unlocked for a long time, and an automatic reboot after 72 hours of inactivity, introduced in iOS 18.1. After a reboot, the phone returns to a state in which most data is encrypted and the keys are not in memory. That makes life much harder for forensic examiners.

The industry fighting this protection has existed for a long time: Cellebrite, Magnet Forensics with its GrayKey product, and others sell data-extraction tools to police. 404 Media has published leaks of these vendors' internal documents before, showing which iPhone models and iOS versions the tools can break. The new leak appears to be from the same series.

A "torture chamber" for LLMs and lawyers with ChatGPT

The second story: someone has created a "torture chamber" for large language models. The announcement does not reveal what the project is, how it works, or what it is for. It could be a test bench for stress tests or red teaming, or something more provocative. It is too early to judge the idea or its usefulness before reading the story.

The third story is clearer. Lawyers, in the authors' words, are "going wild with ChatGPT." The story goes back to 2023, when attorneys in Mata v. Avianca cited court precedents that a chatbot had invented, and the court fined them $5 000. Since then, courts in different countries have regularly recorded similar cases: the model confidently invents citations, and a human does not check them before filing the document.

What the announcement records

• Source: 404 Media podcast, October 8, 2026.
• Stories in the episode: three (iPhone hack, "torture chamber" for LLMs, lawyers and ChatGPT).
• Claim about the iPhone: police are bypassing an important protective feature, based on a leak.
• The named feature, the tool vendor, iPhone models, and iOS versions: not disclosed.
• Details of the LLM project: not disclosed.
• Historical benchmark for lawyers: a $5 000 fine in Mata v. Avianca (2023).
• Benchmark for iPhone protection: reboot after 72 hours of inactivity since iOS 18.1.

Practical takeaways

iPhone owners should keep their systems updated for now: Apple usually closes such bypasses in patches, and forensic tools are tied to specific iOS versions. A long alphanumeric passcode instead of six digits makes brute-forcing noticeably harder, especially if the phone has been rebooted and is in the before-first-unlock state. People at elevated risk (journalists, activists) can enable Lockdown Mode.

For lawyers and companies with legal departments, there is one rule: any reference to a case, statute, or quotation from a language model must be checked against the primary source before it goes to court. It helps to write this into internal policy and keep a log of where AI was used.

For those who follow the AI industry, it makes sense to wait for the full episode: the details of the leak and of the "torture chamber" project will determine how serious these stories are on the merits.