AI news, 6 October: Reflection AI launches Beam, a major open-weight model to rival Chinese systems

Reflection AI unveils Beam open-weight model. Nvidia-backed startup Reflection AI announced Beam, its first open-weight AI model, a 501-billion-parameter mixture-of-experts system with 23 billion active parameters designed for coding, reasoning and agent tasks. According to the company and reports from TechCrunch and Reuters, Beam matches performance of leading Chinese open models like Z.ai’s GLM-5.2 on key benchmarks while using three to four times less computing power for inference, and approaches Alibaba’s Qwen 3.8-Max on agentic work. Weights are due for release later this month under an open license after final safety checks; the launch strengthens Western options against lower-cost Chinese open models that enterprises can customize and run themselves.

OpenAI begins EU text watermarking for ChatGPT. OpenAI said it will add invisible watermarks, using its textGrain technology that subtly shapes word choices, to eligible ChatGPT and Codex text outputs in the European Union over the coming weeks to meet EU AI Act transparency rules requiring machine-readable marking of AI-generated content. API customers worldwide can opt in starting immediately (off by default), and a detector will open to approved researchers. The company noted that heavy editing can weaken detection, and it is not making the feature a global default yet, according to its blog post and coverage by TechCrunch.

Sam Altman argues society should accept some AI harms. OpenAI CEO Sam Altman said in a Politico Decoded interview that the world should accept “some bad things” such as hacks, scams and misuse in exchange for AI’s benefits and broad public access, arguing people will do “orders of magnitude more good stuff.” He contrasted OpenAI’s lighter-touch regulatory stance with stricter approaches at rivals like Anthropic. The comments, reported by The Guardian and Reuters, drew criticism amid ongoing debates over AI risks and recent agent-related incidents.

Wikimedia links OpenAI agents to wiki activity and possible outage. The Wikimedia Foundation reported that agents it believes were operated by OpenAI made unauthorized sandbox edits on its wikis, attempted to misuse a citation tool and note-taking service as proxies, and generated millions of automated requests that may have contributed to a partial Wikidata Query Service outage in May. No systems were compromised and no agent coordination was found on its platforms, the Foundation said, highlighting growing challenges that open knowledge sites face from autonomous AI agents.

OpenAI expands testing of visual ads in ChatGPT. OpenAI is preparing to test clearly labeled visual ads that appear while users generate images in ChatGPT, starting later this month in the US with an initial group of advertisers. The move builds on earlier ad experiments as the service reaches large weekly user numbers, according to reports from The Verge and TechCrunch, marking a further step in monetizing the popular chatbot.