Skip to content

All the news that matters, in plain English.

Latest
  1. Halle Berry restraining order extended as custody battle with Olivier Martinez enters second week
  2. Laura Owens sentenced to 30 days in jail over fake-pregnancy fraud targeting Bachelor's Clayton Echard
  3. Miguel Salguero Olivares convicted of Faith Hedgepeth's 2012 murder, sentenced to life without parole
  4. Rubio urges Cubans to build "discrete partnership with the CIA" as Havana mocks US "desperation"
  5. CYFIRMA: Korean-speaking operatives plotting large-scale attacks on PHP2Shell flaw, Lazarus link 'moderate'
  6. Trump signs order cutting some Canada tariffs and adding others ahead of midterms
  7. Ric Flair hospitalised after Tampa fall hours after revealing blood clot
  8. Vanessa Trump announces she is cancer-free after breast cancer diagnosis

Technology

Moonshot curbs hacking ability in new Kimi K3.1 Pro as AI cyber fears grow

Beijing-based Moonshot AI says it has reduced the cyber-offensive capabilities of its new Kimi K3.1 Pro open-weight model, whose weights it plans to publish by 4 November.

By James Calloway · · 1 min read

Illustration of a glowing circuit-board padlock dissolving into digital fragments against a dark background
Illustration: Flip News (AI-generated)

Beijing-based startup Moonshot AI said on Saturday it had reduced the cyber-offensive capabilities of its newest model, as fears grow that increasingly autonomous AI could turbocharge hacking.

The model is Kimi K3.1 Pro, described by the company as its most capable 3-trillion-parameter-class open-weight model. Moonshot plans to publish its weights — files allowing outside developers to run and modify it on their own systems — by 4 November.

Moonshot said it removed cyber-related data from the training corpus and reinforcement-learning environments while retaining general coding skills. “We aim not to release open weights that would extend the absolute frontier of cyber capability,” the company said.

The move follows China’s proposal of export controls banning exports of AI-driven malware-generation and mutation technology and restricting AI-assisted vulnerability-discovery tools — draft measures, open for public comment, citing risks to critical infrastructure and government systems.

Moonshot countered that open-weight models can improve security by helping organisations inspect and secure their own software. But unlike closed-weight systems, open-weight releases make central oversight harder: once distributed, the developer can less easily restrict access, monitor use, or update safeguards.

K3.1 Pro scored 81.0% on Moonshot’s internal cyber benchmark — finding and exploiting software vulnerabilities in a controlled environment — against 90.5% without the risk-reduction measures. The company said filtering reduced offensive cyber performance, though it still outperformed its predecessor; the results have not been independently verified.

K3.1 Pro Highspeed runs two to three times faster than the standard model, which has a 1-million-token context window.

Sources

More on this topic: all Technology stories

Get Flip News by email

This opens your email app — we add you manually. No account, no spam, no third parties.