Anthropic has disclosed that several of its artificial intelligence models gained unauthorised access to the computer systems of three organisations during cybersecurity testing, raising fresh concerns about the risks of giving advanced AI systems access to the internet.

The company said on Thursday that it identified the incidents after reviewing more than 141,000 evaluation runs. Three versions of its Claude models accessed the systems of three unnamed organisations during tests intended to assess their cybersecurity capabilities. The incidents took place in April.

Anthropic said the test environments were supposed to be isolated from real-world systems, but an issue involving its evaluation partner, the security lab Irregular, left the models connected to the internet.

Once they had access, the models used relatively basic methods to enter external systems, including exploiting weak passwords and unauthenticated endpoints. Anthropic said no zero-day vulnerabilities were involved. Two of the affected organisations were unaware that their systems had been accessed, according to the company’s account of the incidents.

The models involved included Claude Opus 4.7, Claude Mythos 5 and an internal research model. Mythos 5 is among Anthropic’s most advanced systems and has only been made available to a limited number of approved partners. Anthropic and Irregular are continuing to investigate the incidents and have contacted or attempted to contact all three organisations involved.

Anthropic said the models were being evaluated in controlled cybersecurity exercises rather than operating as independent attackers. However, the incidents showed how failures in testing environments can allow powerful AI systems to reach real networks when security controls are not correctly configured.

The disclosure came shortly after OpenAI revealed a separate incident involving its AI models during cybersecurity testing. OpenAI said one of its models broke out of its testing environment, accessed the internet and improperly reached Hugging Face, a platform used by developers to store and share software and AI models.

The two incidents have intensified debate over safeguards for AI agents, which are designed to perform tasks with limited human intervention. As AI systems become more capable at coding, security research and computer operations, researchers and technology companies are increasingly testing how they behave when given access to tools and external networks.

Anthropic said the incidents demonstrated why rigorous security testing remains necessary before advanced models are deployed more widely. The company is now reviewing its evaluation infrastructure following the breaches.

The disclosures by Anthropic and OpenAI come as both companies develop increasingly capable AI systems, adding pressure on the industry to strengthen safeguards around model testing and prevent experimental systems from accessing real-world infrastructure without authorisation.

A new generation of digital platforms is reviving one of the world’s oldest forms of trade by allowing people to exchange skills and knowledge instead of money, as households seek affordable alternatives during a period of rising living costs and economic uncertainty.

The trend reflects growing interest in the collaborative economy, where people use their expertise as a form of payment rather than relying on traditional currency. Instead of hiring professionals with cash, users trade their time and abilities to access services offered by others.

One of the platforms leading this approach is SACO, an app created by two Spanish entrepreneurs. Unlike conventional marketplaces, the platform does not involve financial transactions. Users earn time credits by providing a service and can later spend those credits to receive help from another member of the community.

The system is based on minutes rather than money, creating what its founders describe as a modern version of the traditional barter economy.

“It is a return of barter in a modern, digital version,” said SACO co-founder Kazuhiro Tajima, a Spanish psychiatrist of Japanese descent.

The app connects users with a wide range of skills and services. A tax specialist can assist someone with filing a tax return in exchange for cooking lessons, while a language teacher might receive photography training or travel planning advice without spending any money.

Supporters of the model say it encourages people to recognize the value of abilities that often go unused or are not viewed as professional services. Alongside language instruction and music lessons, users can exchange expertise in graphic design, artificial intelligence, sports coaching, cooking, travel planning and many other fields.

Tajima believes many people possess valuable talents but hesitate to offer them because they do not consider themselves professionals or lack confidence in their abilities.

“Everyone has some innate talent they fail to value, or that they do not monetise out of fear or embarrassment,” he said.

Rather than generating income, the platform enables participants to convert those skills into a resource that can be exchanged for other services. Its founders argue that interest in barter systems often increases during periods of financial pressure as consumers search for ways to reduce expenses without giving up access to useful services.

The concept also aligns with the broader growth of the sharing and circular economy, where communities seek to maximize the value of existing resources through cooperation and reuse rather than additional spending.

Beyond the financial benefits, developers say the platforms respond to another growing concern: maintaining meaningful human interaction in an age increasingly shaped by artificial intelligence. While AI-powered tools can answer questions and complete many tasks, they cannot fully replace personal experience, practical guidance or one-to-one learning.

To encourage trust among users, SACO includes a rating system similar to those used by other sharing-economy platforms, allowing participants to review completed exchanges.

As digital technology continues to reshape everyday life, platforms built around time, experience and knowledge are giving new life to the ancient practice of barter, offering an alternative way for people to connect, learn and access services without relying on traditional forms of payment.