Bitcoin heist, AI under scrutiny and new drones at sea
A huge crypto theft, fresh questions about who reads your AI chats, and a busy day for new hardware.
The TechMap Live reporter · · 2 min read · from 10 reports
Listen to this briefing
About 1 min to listen
Read the transcript
Let's catch you up on what's been happening in tech. The biggest story is a crypto theft: hackers pulled about four thousand Bitcoin, worth roughly three hundred and twenty million dollars, out of Liquid Network, a Bitcoin sidechain run by the Canadian company Blockstream. Security was a theme all round, actually. Researchers found a serious Bluetooth flaw in Skullcandy's Dime 3 headphones that lets strangers pair with them, and there's no fix yet. Over in AI, OpenAI admitted that outside contractors are reading real ChatGPT conversations to help improve the model, which is worth knowing before you type anything sensitive. Microsoft AI, meanwhile, put out its first draft code of conduct, saying its models must stay under human control and can't refuse to be switched off. And a lawyer in New Mexico got sanctioned after ChatGPT invented testimony and witnesses in a murder appeal filing. On the hardware side, German firm Quantum Systems launched its first sea drones, MediaTek has a new flagship phone chip built for on-device AI, Fujitsu open-sourced a quantum software toolkit, and a fifteen-year-old in Canada built a turtle-shaped robot that detects coral bleaching. Busy day.
A very bad day for crypto
The biggest story by far is a theft from Liquid Network, a Bitcoin sidechain built by a Canadian company called Blockstream. Hackers drained roughly four thousand Bitcoin from its shared wallet, worth about three hundred and twenty million dollars. That makes it one of the largest crypto thefts in a while.
It wasn't the only security headache. Researchers found a critical Bluetooth flaw in Skullcandy's Dime 3 headphones that lets strangers pair with them without permission, and there's no fix available yet. On the more hopeful side, a security firm called Entrust added new tools to its cryptographic platform, designed to turn lists of software encryption components into something companies can actually act on.
AI's trust problem
OpenAI confirmed that outside contractors are reading and evaluating real ChatGPT conversations to improve the model. The company says it's for annotation, but it's a reminder that what you type into a chatbot may be seen by human eyes.
Meanwhile Microsoft AI published its first draft code of conduct, saying its models must stay under human control and can't resist being shut down or expand their own tasks. And in New Mexico, the state supreme court sanctioned a lawyer who filed a brief full of made-up testimony and witnesses that ChatGPT had hallucinated for a murder appeal. The pattern is clear: the tools are powerful, and the guardrails are still being written.
New hardware, on sea and in chips
German drone maker Quantum Systems moved into the water with its first marine lineup. Seavex is a surface drone, Subvex goes underwater, and together they're built for maritime reconnaissance, with up to two months of autonomy and dives as deep as six kilometres.
Elsewhere, Taiwan's MediaTek launched the Dimensity 9600 Pro, a high-end phone chip aimed at running AI directly on the device. Fujitsu released OpenQARP, a free open-source package with over a hundred quantum algorithm functions for researchers. And a fifteen-year-old in Canada built an underwater robot shaped like a sea turtle that spotted coral bleaching with ninety-six percent accuracy in a test.
Questions people ask
What happened to Liquid Network?
Hackers drained roughly four thousand Bitcoin, worth about three hundred and twenty million dollars, from its federated wallet. Liquid Network is a Bitcoin sidechain developed by the Canadian firm Blockstream.
Is someone reading my ChatGPT conversations?
Possibly. OpenAI confirmed that external contractors read and evaluate user conversations to improve the model, as part of a data annotation process.
What did Microsoft AI's code of conduct say?
It's a first draft requiring that AI models remain under human control. The models must not resist shutdown or expand their tasks on their own.
The events in this briefing
Every event summarised above, in the order it was reported. Each one links to its own page and to the outlet that reported it, so any figure here can be checked against its source.
- Quantum Systems launches Seavex and Subvex marine drones · dev.ua
- MediaTek launches Dimensity 9600 Pro high‑end mobile chip · Xataka
- Fujitsu releases OpenQARP as open-source quantum application package · The Quantum Insider
- OpenAI contractors reading ChatGPT user prompts · Golem.de
- Canadian teen builds underwater robot resembling sea turtle · Zoomit
- Hackers steal $320M in Bitcoin from Liquid Network · ITC.ua
- Critical Bluetooth vulnerability found in Skullcandy Dime 3 headphones · ShiftDelete
- New Mexico Supreme Court Sanctions Lawyer for ChatGPT-Hallucinated Citations · Habr News
- Entrust Expands Cryptographic Security Platform with CBOM Action Capabilities · The Quantum Insider
- Microsoft AI Publishes Draft Code of Conduct Requiring Human Control · iThome Taiwan
This briefing was written by an automated reporter from 10 event reports already published on this site. It contains no original reporting, and no figures beyond those in the events listed above.
More briefings
- Robotics surge and new tech launches as September events stack up
- Tech map update: payment shifts, AI labs, and cyber alerts
- Tech update: AI defence, phone teardowns and security alerts
- Tech updates: Apple foldable yield, crypto theft, and security patches
- AI security tests and software updates shape tech week