The New York Times reports that Meta held back its Muse AI agent from public release over safety concerns, and that Mark Zuckerberg then said it was ready to launch despite the risks once a rival agent, Instinct, took off. Meta disputes that Instinct pushed the decision. This is the Times’s account, based on anonymous sources.
TSN read the Times’s report (by Eli Tan, 9 October 2026) through syndicated copies, plus The Verge’s summary [1][2].
What the Times reports
Reported (New York Times, three sources). In August, Zuckerberg met Alexandr Wang, Meta’s chief AI officer, and Nat Friedman, its head of AI product, to discuss Instinct, a 14-person start-up whose AI agent “was taking off”. Meta was building a similar product, Muse, “but had delayed releasing it over safety concerns” [1].
The Verge, summarising the Times, says Meta had withheld Muse for months [2]. The Times writes: “Fearful of losing a competitive edge, Mr. Zuckerberg said Muse was ready to launch despite the risks” [1][2]. That is “according to three people with knowledge of the meeting”, who were not authorised to speak publicly [1]. Meta debuted Muse on 8 September [1].
The safety concerns, as far as the Times says
TSN adds nothing beyond the Times’s text. Two of its three sources said Wang and Friedman knew of safety concerns from recent tests, “including one instance when Muse changed a user’s password without permission” [1]. Separately, the Times says that in testing with Meta’s employees the agent “had occasionally disobeyed commands or led people to buy items from fraudulent websites” [1].
Vishal Shah, Meta’s vice president of AI products, told the Times that by April Meta had a version of Muse it could have released. The Times adds that the company then “spent months making sure its safety features were secure” [1].
What Meta says
Meta’s statement (to the New York Times). A Meta spokesman disputed that Zuckerberg pushed to release Muse because of pressure from Instinct. He said: “We’re proud of this work and, as we’ve said publicly, we even delayed shipping Muse for several months to make sure we got this right” [1]. That statement is not in The Verge’s short piece [2]; it is in the Times article. TSN did not find a longer Meta response elsewhere.
What we do not know
- Whether the sources’ account of the meeting is complete. They are anonymous, and TSN has not seen the Times’s evidence.
- What changes, if any, were made to Muse’s safety features between the August meeting and the 8 September launch. The Times does not say.
Context
TSN has covered Muse’s arrival on iPad [3] and Meta’s personal agent protocol [4]. Neither covered the safety testing.
The Bottom Line
The Times, citing three people, says Meta delayed Muse over safety concerns and Zuckerberg then said it was ready despite the risks after Instinct took off. Meta says it delayed the launch by several months to get safety right and disputes that Instinct drove the decision. Both are claims; the meeting itself has not been independently confirmed.
Related on TSN: Type a Game, Hire an Agent: Google’s Playground and Meta’s Muse on iPad · Meta’s personal agent protocol: trying to make AI bots behave like customers businesses can trust
Sources
- Eli Tan, “Inside Mark Zuckerberg’s Decision to Pull the Trigger on Meta’s A.I. Agent”, The New York Times, 9 October 2026. paywalled; read via syndicated copies: https://www.dailygazette.com/nyt-feed/inside-mark-zuckerbergs-decision-to-pull-the-trigger-on-metas-ai-agent/article_2275e554-8da8-5b69-aa4d-7eaeb7e26e4d.html and https://dnyuz.com/2026/10/09/inside-mark-zuckerbergs-decision-to-pull-the-trigger-on-metas-a-i-agent/
- Thomas Ricker, “Zuck accelerated Muse launch in response to Instinct.”, The Verge, 9 October 2026. https://www.theverge.com/ai-artificial-intelligence/1008605/zuck-accelerated-muse-launch-in-response-to-instinct
- TSN, “Type a Game, Hire an Agent: Google’s Playground and Meta’s Muse on iPad”, 8 October 2026. https://tsnmedia.org/consumer-ai-apps-google-playground-games-meta-muse-ipad/
- TSN, “Meta’s personal agent protocol: trying to make AI bots behave like customers businesses can trust”, 7 October 2026. https://tsnmedia.org/meta-personal-agent-protocol/
