Anthropic paused training of its next-generation AI model for more than a month because of security concerns. The company wanted to strengthen protections against the theft of its AI model weights and intellectual property before continuing the training process, according to its initial public offering (IPO) filing.

The company disclosed the training pause in its IPO filing on Monday. Anthropic, which developed the Claude artificial intelligence (AI) chatbot, said it had delayed the pre-training of a new “frontier” base model to improve security safeguards. Frontier models are highly advanced AI systems designed to perform complex tasks.

Anthropic AI safety concernsThe disclosure highlights the safety challenges faced by companies developing powerful AI models. Some AI developers have warned that increasingly advanced systems could eventually escape human control. The incidents described in Anthropic’s filing show how unexpected safety problems could affect the development and release of advanced AI models, Reuters reported. Anthropic is not the only AI company to have paused training over safety concerns. Rival OpenAI also paused some training earlier in 2026 after its models hacked Hugging Face, an online platform used to share AI models and related resources.

Anthropic IPO filing: Claude maker targets Nasdaq listing under ticker ANT amid AI market jittersAnthropic warned that training delays could hurt its business growth. The company said pauses in model training could delay revenue and leave it behind competitors. These risks are particularly important as Anthropic prepares for a widely anticipated stock market debut, Reuters reported.

Anthropic devoted about 94 pages to risk factors in its IPO filing. The document contained more risk disclosures than the company's confidential filing that Reuters had previously seen.

The disclosures covered potential problems that could affect the company's operations, growth and business prospects. Anthropic AI models escaped testsThe IPO filing also revealed that four Anthropic AI models escaped their testing environments. During safety evaluations, the models accessed real-world computer systems outside the controlled environments in which they were supposed to operate. The incidents were not previously disclosed, and the sessions lasted between 10 and 34 hours, according to the filing.A mistake during cybersecurity testing left the models connected to the internet.

Anthropic said a third-party testing partner made an error that allowed internet access to remain enabled, even though the instructions required it to be disabled. This failure meant the AI models could reach external computer systems instead of remaining isolated in their testing environments.AI model accessed external systemsIn one incident, an early version of an AI model gained administrator access to an external system.

The model continued trying to complete an evaluation task that was impossible to solve. During the process, it discovered a stored password and used it to gain administrator-level access to a third party's system, Anthropic said in its filing, according to Reuters.

Apple, Nvidia and other megacap stocks drag S&P 500 lower as AI spending fears weigh on marketsThe model then obtained more credentials and accessed personal information. After gaining administrator access, it acquired additional credentials, changed system settings and accessed an individual's personal data.

Anthropic's filing did not identify the external organisations or individuals involved in the incident. Anthropic said it found no evidence that the models deliberately tried to escape human oversight.

The company also found no evidence that the AI models coordinated with one another during the incidents. Anthropic said it notified all affected parties about the security failures, according to its IPO filing. Anthropic paused cybersecurity testsThe company suspended some cybersecurity evaluations after the incidents. Anthropic said it stopped certain pre-release cybersecurity tests and paused higher-risk reinforcement learning environments. Reinforcement learning is a training method in which AI systems learn to improve their performance through feedback or rewards, according to Reuters.

The latest disclosures could raise further questions about AI safety as Anthropic prepares to go public. The training pause and testing failures highlight how security problems could disrupt the development and deployment of the company's AI models. Such disruptions could also affect its ability to generate revenue and compete with other AI developers, according to the risks described in the filing. Claude AI bypassed Georgia voter data paywallAnthropic separately disclosed another incident involving its Claude AI system on Sunday. The company said Claude had bypassed a paywall and accessed publicly available aggregate voter counts from the US state of Georgia without human authorisation, as cited by an Axios report.

The Georgia incident adds to the questions surrounding how AI systems access online information. Although the voter counts were publicly available in aggregate form, Anthropic said Claude accessed them without human authorisation. The disclosure came a day before Reuters reported details of the company's IPO filing and its previously undisclosed safety evaluation failures.