Zuckerberg says Meta has reached AI ‘frontier’

Meta Platforms chief executive Mark Zuckerberg has declared that the company has caught up with rivals at the “frontier” of artificial intelligence, vowing to sustain the heavy spending that has defined its AI push and rejecting calls from competitors for an industry-wide slowdown. The statement accompanied the release of Muse Wonder, a new AI model that Meta claims outperforms comparable systems from Anthropic and OpenAI on several technical benchmarks, marking a significant escalation in the race among the largest US technology firms.

The release represents a milestone for Zuckerberg, who since last year has committed billions of dollars to overhaul Meta’s AI leadership, construct vast data centres, and recruit top researchers on multimillion-dollar salaries. That investment has begun to show commercial returns. The Muse personal AI assistant, which uses autonomous bots to handle everyday tasks such as grocery ordering and travel booking, has become the most downloaded app on both Apple and Android app stores in the United States, and Meta’s share price has risen nearly 10 per cent since the assistant launched in early September. The company now intends to integrate Muse Wonder into the app, with the model’s most advanced versions sold directly to businesses, placing it in direct competition with Anthropic and OpenAI for corporate AI contracts.

Zuckerberg said the new model was “going to allow much more sophisticated behaviours that are at the frontier of what is being done in the AI field,” and signalled that further investment is coming. Meta is already working on the next iteration, he said, and model development would “probably feel like it’ll go quite a bit faster from here.” The company said Muse Wonder can handle complex tasks ranging from interpreting a lease to making sense of financial or health information, positioning it as a tool for enterprise customers as well as consumers.

The decision to keep the best versions of Muse Wonder closed and sell them directly to businesses places Meta in direct competition with Anthropic and OpenAI for corporate AI contracts, a serious test of whether its technical progress can translate into commercial scale. The company’s strategy also reflects a deliberate philosophical divergence from its rivals. Zuckerberg has repeatedly rejected proposals by Anthropic’s Dario Amodei and OpenAI’s Sam Altman for a coordinated slowdown in AI development, and he played a central role in shaping the White House “Accord on Super Intelligence” under President Donald Trump, a self-regulatory framework that critics argue falls well short of a formal global governance regime.

Zuckerberg dismissed concerns that the technology is advancing beyond human control, declining to offer his own estimate of the probability that AI leads to catastrophic outcomes, a figure known in Silicon Valley as “P(doom)”. He argued instead that responsibility for safety should rest with individual labs and that AI can be engineered with normal guardrails, rejecting what he called the “anthropomorphic vision” of AI as an autonomous force. This position puts him directly at odds with the heads of rival labs, including Anthropic’s Dario Amodei and OpenAI’s Sam Altman, both of whom have urged a slower development pace.

The strategic divergence is stark. Meta is pushing a vision of “personal superintelligence” aimed at individual consumers, while also selling its best Muse Wonder versions directly to businesses, placing it in direct competition with Anthropic and OpenAI for corporate contracts. The closed nature of the model, with premium versions reserved for paying enterprise clients, signals that Meta intends to monetise its frontier research aggressively rather than release it openly. Zuckerberg has also played a central role in shaping the White House “Accord on Super Intelligence,” a self-regulatory framework that critics argue falls short of formal global governance.

On the question of existential risk, Zuckerberg declined to offer his own estimate of the probability that AI leads to mass destruction, a metric known in Silicon Valley as P(doom). Instead, he argued that responsibility for safety should rest with individual laboratories, describing AI as a technology that can be engineered with normal guardrails rather than a force that will act independently. That stance puts him in direct opposition to Anthropic’s Dario Amodei and OpenAI’s Sam Altman, who have both urged an industry-wide pause in development. Zuckerberg has instead championed a vision of “personal superintelligence” for individual consumers, positioning Meta’s assistant-led approach against the enterprise-focused strategies of its main rivals.

The commercial stakes are considerable. Muse Wonder will be sold as a closed model, with its best versions offered directly to businesses, placing Meta in direct competition with Anthropic and OpenAI for corporate AI contracts. Zuckerberg signalled that spending will continue, saying the company is already working on the next step and that development would “probably feel like it’ll go quite a bit faster from here.”

His dismissive stance on safety concerns marks a clear philosophical split within the industry. While executives at Anthropic and OpenAI have called for collective caution, Zuckerberg argues that responsibility should rest with individual labs and that AI can be engineered with normal guardrails. He helped craft the White House “Accord on Super Intelligence” under President Donald Trump, a self-regulatory framework that critics say falls far short of formal global governance.

The divergence is more than rhetorical. By selling its best models directly to businesses and pushing a vision of “personal superintelligence” for consumers, Meta is positioning itself as a commercial counterweight to the safety-first posture of its rivals. The question is whether self-regulation, pursued at speed, can keep pace with a technology that Meta itself acknowledges is advancing faster than ever. For investors, the near-term answer is encouraging; for regulators and the public, the risk calculus remains open.