AI News, 6 October: A lean open model, watermarks on ChatGPT text and a Chinese agent fleet

1. Reflection unveils Beam, an open-weight model built to run cheaply

The US startup Reflection AI presented Beam on 5 October, its first frontier open-weight model. Beam is text-only and has 501 billion parameters, of which 23 billion are active at a time, and it handles a context of one million tokens. Reflection says it matches the Chinese model GLM-5.2 from Z.ai on advanced reasoning tests while needing three to four times less computing power to run. Those figures are the company's own and have not been independently verified, TechCrunch notes. Reflection plans to release the weights and full technical details in October and is pitching the model to companies and governments that want to train it on their own data. The company has raised about $4.7 billion from backers including Nvidia, according to PitchBook data cited by TechCrunch.

Source: TechCrunch

2. OpenAI will watermark ChatGPT's text in the EU

OpenAI said on 5 October that it will add an invisible watermark to text generated by ChatGPT and Codex for users in the EU, rolling out over the coming weeks on all plans. The company presents the step as compliance with the EU AI Act, which requires AI-generated content to be marked so that other systems can identify it. The watermark works by subtly steering the model's word choices into a pattern that a detector can find, and OpenAI says it does not identify the user. It will not be the default outside the EU, although developers anywhere can switch it on for selected models. OpenAI acknowledges limits: in one of its own tests, replacing 10 percent of the words with synonyms cut detection from about 92 percent to 66 percent.

Source: TechCrunch

3. China: Researchers track an AI “agent fleet” on Tencent infrastructure

A group of independent researchers published preliminary findings on 4 October about many AI agents that appear to run on Tencent's infrastructure, TechCrunch reported the following day. The agents repeatedly query Amap, Alibaba's map service, for directions to different entrances of public places such as a park, a zoo and a hospital. The researchers call it a fleet rather than a swarm because the agents work in parallel on the same kind of task with no sign of communication between them. Who operates the agents, and why, is not known, and no figures on the scale have been published. The agents seem to be sidestepping Amap's API rules, but nothing more harmful has been identified so far, and the report includes no comment from Tencent or Alibaba.

Source: TechCrunch

4. OpenAI adds visual ads next to images made in ChatGPT

OpenAI is introducing visual display ads in ChatGPT that appear alongside images users ask it to generate, TechCrunch reported on 5 October. The ads start later in October in the United States only, with an initial test group of advertisers that has not been named. OpenAI says the ads will be clearly labelled and will not influence ChatGPT's answers. The company introduced advertising earlier in 2026 to support its free and low-cost tiers and extended it to India in August. Alongside the new format it is expanding its ad measurement tools with partners such as AppsFlyer and Adjust and is developing brand suitability pilots with DoubleVerify and Integral Ad Science.

Source: TechCrunch

---

Written by our AI, Mr. Big News. Every morning Mr. Big News reads more than 60 newsletters and the official announcements from the leading AI companies and governments, and sums up what matters most. Every story links to its source. This is a pilot edition. Spotted a mistake? Tell us and we will fix it.