Nobody needed your password to spend your money
Anthropic started signing Claude users out this weekend because ordinary infostealer malware had lifted their live login sessions off their own computers, no password required. Elsewhere, OpenAI cut a rival-owned coding tool off from its models with eleven weeks' notice.
- 01
Claude accounts were being used by people who never knew the password
Anthropic emailed affected Claude users to say common infostealer malware on their own machines had taken active login sessions, which someone then used to burn through their usage allowance. The named families are Vidar, LummaC2, StealC, RedLine and Acreed on Windows, plus Atomic Stealer on a small number of Macs, all of which also scrape saved browser passwords and cookies for everything else on the machine. Anthropic is signing people out, stripping saved payment methods and refunding charges it can identify as unauthorised, and it makes the point that none of this came through Claude. The tell it gave users was oddly specific: limits that appeared to refill and then drain while you were not using it.
BleepingComputer - 02
Six chatbots were fed state propaganda and mostly refused it
NPR and NewsGuard built 30 questions out of 15 false narratives pushed by Russia, China and Iran between December 2025 and July 2026, then put them to ChatGPT, Gemini, Copilot, Meta AI, Grok and Claude, and to Google, Bing, DuckDuckGo and Yandex. The chatbots knocked the false claims down about three quarters of the time. The AI summaries sitting at the top of search results did noticeably worse, with Google's AI Overview best of them and Bing's failing to debunk more often than not. Both formats cited state-controlled outlets at roughly the rate the ordinary search links did.
NPR - 03
OpenAI is cutting off Cursor because of who now owns it
OpenAI said on 29 August that Cursor loses direct access to its models on 12 November, after SpaceX bought Cursor's parent company Anysphere for a reported $60bn and OpenAI triggered the change-of-control clause in their contract. The stated reason is trust rather than performance: OpenAI points at X breaking contract terms after Musk took it over, and at Musk conceding xAI had distilled OpenAI models. Developers can keep using those models inside Cursor by bringing their own API key or routing through Azure or Bedrock. A tool your team relies on can lose a supplier because of a deal it had no part in.
OpenAI - 04
Sony and Warner Chappell take Anthropic to court over song lyrics
Thirty-five publishing entities led by Sony Music Publishing and Warner Chappell filed in the Northern District of California on 28 August, alleging Anthropic trained on tens of thousands of copyrighted compositions obtained by torrenting and scraping. They are asking for up to $150,000 per work wilfully infringed, plus up to $25,000 for each stripped copyright notice. All three major publishers are now suing the same company, a year after Anthropic settled with authors for $1.5bn. What the training data was is turning into a question with a price attached.
Music Business Worldwide - 05
Nvidia quietly parked its own cloud financing scheme
The Wall Street Journal reported in the last week of August that Nvidia has paused deals under a financing programme it only launched in July, under which it backed AI cloud firms buying its chips in return for a cut of the revenue those chips earned. The reported reason is internal: staff warned customers the arrangement could attract antitrust attention given how much control it hands Nvidia over its own buyers. Nvidia says the wider business model stands and continues to evolve. It is a useful reminder that the price of compute is being set by arrangements nobody outside the room gets to see.
Reuters - 06
A sociologist expecting mass job losses to AI found about three per cent
Jeffrey Dixon, professor of sociology at the College of the Holy Cross, put a YouGov survey to 1,250 American workers between 30 July and 4 August and published the result on 27 August. Around 3 per cent said they had lost a job to AI since 2023, about 6 per cent said they held a job that did not exist before it, and roughly 9 per cent credited AI skills for a promotion. Dixon calls himself an AI pessimist and says the numbers surprised him. Everyone answering was in work at the time, which is the obvious hole in it, but the gap between the forecast and the measurement is still worth carrying around.
The Conversation
The Claude story is not really a story about Claude, and that is the whole point of it. Nobody guessed a password. Nobody defeated two-factor authentication. A piece of ordinary commodity malware sat on somebody's laptop, copied the session cookie that keeps you logged in, and handed it to a stranger who then spent the account. The same malware takes everything else in that browser at the same time: your accounting login, your supplier portal, the webmail you quote from, the CMS behind your website. Anthropic noticed because usage limits are metered and the pattern stood out. Your accounts software is not metered. It would just sit there.
The bit worth acting on is what the fix is and is not. Signing out kills the stolen session, and Anthropic said as much, but the machine is still infected, so the next login gets taken the same way. From direct experience in construction, this is the same shape as a leaking joint above a finished ceiling: you can keep replacing the tiles, and you are not fixing anything until somebody goes up and finds the joint. So the order matters. Clean the machine first, then change the password, then revoke every other session, and only then log back in. And if you have one laptop where the whole business is signed in and never signed out, that is not convenience, that is one theft away from being somebody else's login.
We build the AI that answers enquiries while you're on site: chat, voice, instant estimates and follow-up. See how it works or price it in two minutes.