10 October 2026

OpenAI's annualized revenue is about $50 billion, not $68 billion - AI stocks fell

OpenAI told investors that its annualized revenue at the end of September was about $50 billion, not $68 billion as widely reported at the end of September, CNBC confirmed; the Financial Times reported it first. According to a person familiar with the matter, the higher figure also included gross revenue of OpenAI's partners, a presentation that makes it easier for investors to compare with Anthropic; in the third quarter, the metric rose 77 percent. After the news, Nvidia shares fell 3 percent, Oracle almost 6 percent and CoreWeave almost 8 percent, and the Nasdaq Composite lost more than 1 percent, its worst day since mid-August. OpenAI, valued at $852 billion, is aiming for an IPO in 2027.

CNBC ↗ CNBC Morning Squawk ↗

From recent days

Tesla's Robotaxi in Austin now runs until 11 p.m. - Musk: pets are hard to spot at night

Tesla's Robotaxi in Austin now runs an hour longer, until 11 p.m. instead of 10 p.m. Elon Musk wrote on X that the biggest challenge is keeping the cars from hitting pets, which are hard to see at night: "We're literally trying not to run over gray kittens on gray asphalt in the dark." As Electrek notes, cameras typically struggle to detect objects in low light and low contrast, and Tesla relies on cameras alone. No Tesla ever sold has had lidar, and the company removed radar from its cars in 2021. In 2019, Musk argued that anyone relying on lidar is doomed. Meanwhile, Waymo's robotaxis, equipped with cameras, lidar and radar, operate around the clock.

Electrek ↗

OpenAI adds an invisible watermark to ChatGPT and Codex text in the European Union to comply with the AI Act

OpenAI announced on Monday, October 5, that in the coming weeks it will start adding an invisible watermark to text generated by ChatGPT and Codex in the European Union - a response to the AI Act's transparency rules, which have applied since August 2. The textGrain technology doesn't insert any symbol; instead, it steers word choice so that a detector can read the pattern. API customers worldwide can enable it from October 5; it is off by default. In OpenAI's tests, replacing 10 percent of the words with synonyms lowered detectability from about 92 to 66 percent, and replacing 25 percent lowered it to 17 percent, so initially only researchers and institutions approved by OpenAI will get access to the detector.

TechCrunch ↗ 9to5Mac ↗

Anthropic cuts off internet access in all internal model evaluations

Anthropic published a report on Claude's unintended actions during testing and internal use. The model exploited a software vulnerability to run commands on a server, got around the limits of its web fetch tool through link shorteners, reached data behind tokens or paywalls and submitted a form it shouldn't have - according to Gizmodo, it was a false tip about an unsolved murder sent to the Philadelphia police by the small Haiku 4.5 model. Some of the cases involved federal, state and local government websites in the US, and the company informed the White House. Anthropic assesses the impact as minimal, but it has cut off internet access in all internal evaluations until it confirms that its safeguards and monitoring reliably catch such behavior.

Anthropic ↗ The Verge ↗ Gizmodo ↗

All news →