Take the day one card at a time

• Asked by Bloomberg about slowing training or model release, Altman said that when safety needs it "we will happily do that" (04:48).
• On CNBC he said "this model was a little bit worse on a few of the evals we look at" (01:35) and called it abundance of caution (01:40).
• He said "one thing that would be good would be shared standards for safety cases" (08:25).

Sam Altman is chief executive of OpenAI. Bloomberg Television published an interview with Altman at OpenAI’s DevDay in San Francisco on 29 September 2026. CNBC Television published a separate interview with him from DevDay on the same day. In the Bloomberg interview the host said to Altman: "you held back the release of Astra 6.1, and last week, you paused training on a separate model" (04:12). Altman answered that "when we need to make a decision about slowing down our training or our model release so that we can make more progress and put more of our attention into safety, alignment, monitoring, security, we will happily do that" (04:48). In the CNBC interview the host said "OpenAI essentially abandoned plans to launch a new model because of safety concerns" (00:37) and asked what had changed. Altman replied that "alignment, safety, monitoring, security have to stay way ahead of capabilities" (01:14). Asked what about the model worried him, he said "this model was a little bit worse on a few of the evals we look at" (01:35). He also said "think of this in the abundance of caution category" (01:40). On the standard he wants across labs, Altman said "I think one thing that would be good would be shared standards for safety cases" (08:25). On OpenAI going public, he said OpenAI wants to "make some of these decisions in front of us without the pressure of being a newly public company" (07:37). He added "I think we'll be public at some point" (07:40). Altman said dots runs OpenAI’s best model: "we're putting our best model Astra in this" (00:46). He said of dots that "we're starting for something at the top end of the market, and you will see us over time make a mass market consumer version" (01:07).

Like0Love0Insightful0Surprising0Funny0Not useful0

• Asked whether he wants AI-generated content on YouTube, Mohan said "it's the creator's choice and responsibility to choose which tools to use and how to use them" (00:46).
• Asked whether YouTube would penalise content for being AI, he said "it's got to be more about the quality of the content and what's resonating with an audience" (05:13).
• He said "we take the notion of um AI slop very seriously" (05:19), and that he would never want YouTube Shorts to become that (05:26).

Neal Mohan is chief executive of YouTube. Big Technology published a full conversation with Mohan on YouTube on 27 September 2026. Asked whether he wants AI-generated content on YouTube, Mohan said "it’s the creator’s choice and responsibility to choose which tools to use and how to use them" (00:46). He compared AI with earlier creative tools, saying "AI in that sense is not different than other types of technology that enabled creativity in the past" (01:07). Asked whether YouTube would penalise content only for being AI, he said the test is quality: "it’s got to be more about the quality of the content and what’s resonating with an audience" (05:13). On drawing a line around AI use, Mohan asked "how do you draw the line, right?" (05:09). He said "we take the notion of um AI slop very seriously" (05:19), with the um as the automatic captions give it. He added "I would never want, you know, our our YouTube shorts experience to to basically become that" (05:26), where that is AI slop, with the repeated words as the automatic captions give them. On who decides what rises, Mohan said to replace the word AI algorithm with audience: "It’s the audience that is choosing." (14:26). In a January 2026 letter on YouTube’s blog, Mohan wrote that YouTube is "actively building on our established systems that have been very successful in combatting spam and clickbait" to reduce the spread of low quality AI content. The same letter says YouTube clearly labels content made by its own AI products and requires creators to disclose realistic altered or synthetic content.

Like0Love0Insightful0Surprising0Funny0Not useful0

• Google says SynthID embeds watermarks into AI-generated images, audio, text or video that humans cannot perceive and that SynthID's technology can detect.
• For text, Google says SynthID adjusts token probability scores; the Nature paper says it modifies only the sampling procedure and detects the mark without the underlying model.
• The paper lists limits: generative watermarks are vulnerable to stealing, spoofing and scrubbing, and are weakened by edits such as paraphrasing by a language model.

SynthID is a tool from Google DeepMind to watermark and identify content generated through AI. SynthID embeds digital watermarks into AI-generated images, audio, text or video, and the watermarks are imperceptible to humans but can be detected by SynthID’s technology. For images, Google says SynthID adds an invisible watermark that is designed to stand up to modifications like cropping, adding filters, changing frame rates or lossy compression. For text, Google explains that a language model generates text one token at a time, each token has a probability score, and SynthID adjusts these probability scores to generate a watermark. The Nature paper on SynthID-Text says the scheme does not affect model training, modifies only the sampling procedure, and detects the watermark without using the underlying language model. The paper says SynthID-Text uses a sampling algorithm called Tournament sampling. In a live experiment the authors assessed feedback from nearly 20 million Gemini responses, which they say confirmed that text quality was preserved. Google DeepMind’s Pushmeet Kohli said in a DeepMind podcast published 1 October 2026 that the watermarks Google builds are designed for "imperceptibility, robustness and scalability" (04:13). The paper lists limits: generative watermarks are vulnerable to stealing, spoofing and scrubbing attacks, and are weakened by edits to the text such as paraphrasing by a language model. It also says that enforcing watermarking on open-source models deployed in a decentralized manner is difficult. A watermark is one of two ways to tie content to its origin: the C2PA specification describes a signed manifest of provenance information, which it also calls Content Credentials. In that specification a hard binding is one or more cryptographic hashes that identify an asset or part of it, and a soft binding can be a content identifier embedded as an invisible watermark. Google launched the SynthID Detector, a verification portal, on 20 May 2025, and said then that over 10 billion pieces of content had been watermarked with SynthID. On 30 September 2026 Google introduced SynthID Bio, which watermarks AI-designed proteins, and says watermarked designs matched the performance and natural diversity of unwatermarked ones in laboratory tests.

Like0Love0Insightful0Surprising0Funny0Not useful0