Anthropic Mythos vs OpenAI: Cybersecurity Becomes AI Arena

Suswati Basu is a multilingual, award-winning editor. She was shortlisted for the Guardian Mary Stott Prize and longlisted for the Guardian International Development Journalism Award....

Top AI models are vying for dominance on a new battlefield: cybersecurity, the invisible infrastructure that keeps the digital world running. The tech giants OpenAI and Anthropic have spent the past week playing a game of one-upmanship, as both entities launched more secure AI systems.

Anthropic's newly unveiled Claude Mythos Preview gives us a glimpse of what that process might look like. While the model is generally trained for coding, it has demonstrated unexpected strength in cybersecurity tasks. As described in Anthropic's own materials, "the model that we're experimenting with is by and large as good as a professional human at identifying bugs."

"These models have capabilities which are raising the bar from a cybersecurity point of view with their ability to help defenders as well as potentially help adversaries," Cisco's SVP & Chief Security & Trust Officer Anthony Grieco says in a video statement.

In a blog post, the company explains, "Mythos Preview demonstrated the ability to independently identify, then chain together, a set of vulnerabilities," referring to the process of linking multiple small flaws into a larger, more impactful exploit. What is usually a manual process for experienced researchers is now essentially automated.

To manage the risks, Anthropic has launched Project Glasswing, a collaboration with 40 organizations responsible for critical infrastructure -- an approach that OpenAI cyber researcher Fouad Matin has criticized, saying, "no one should be in the business of picking winners and losers."

OpenAI Takes Pot Shot at Anthropic Over its Closed AI Network Of Testers For Mythos

OpenAI's latest initiative, detailed in its "Trusted access for the next era of cyber defense" post, leans into controlled deployment. The company is experimenting with restricted-access models for cybersecurity professionals in a bid to "empower defenders by giving broad access to frontier capabilities." However, the tech firm was quick to talk about "democratized access" and warned against "arbitrarily deciding who gets access."

Reporting from The Verge also points to growing pressure between leading AI labs. Internal discussions highlight the speed at which capabilities are advancing and the dwindling margin for error. OpenAI's chief revenue officer, Denise Dresser, reportedly told staff in a memo that Anthropic's story is "built on fear, restriction, and the idea that a small group of elites should control AI."

Anthropic, which has taken a safety-first stance on AI, says that "software ate the world." That concern was echoed this week by David Solomon, chief executive of Goldman Sachs, who said the bank is "hyper-aware" of the risks posed by Anthropic's Mythos model. The firm is said to be working closely with Anthropic and cybersecurity partners to assess potential threats, even as Solomon and other big American bankers were summoned to Washington last week to discuss the Mythos model.

Nearly every aspect of modern life depends on code. If that code is vulnerable, so is everything built on top of it.

Also in Tech News

Amazon, Apple, and Globalstar Signal a New Era of Hybrid Networks

Amazon is the latest tech giant moving deeper into satellite infrastructure, announcing plans to acquire Goldstar and partner with Apple. Under its low Earth orbit system, the company says it will "extend cellular coverage to customers beyond the reach of terrestrial networks." It is not expected to replace traditional telecom infrastructure, however.

Globalstar brings its own spectrum, satellites, and operational expertise into Amazon's orbit. "There are billions of customers... beyond the reach of existing networks," said Amazon's devices chief, Panos Panay, adding that the goal is to "bridge that divide" with "faster, more reliable service in more places."

Apple's own ecosystem will add a consumer layer, with features like Emergency SOS. Meanwhile, the multinational company will help power current and future capabilities across iPhone and Apple Watch.

It appears that connectivity is becoming hybrid, and the infrastructure supporting it will no longer just be earthbound.

China Draws a Line on Humanlike AI Interactions

China's top internet regulator has issued new guidance against AI anthropomorphism. Authorities from the Cyberspace Administration of China warned against the design and deployment of AI systems that blur the boundary between machine and human identity.

The regulator specifically called for clearer labeling of AI-generated content and stricter controls on systems that imitate human interaction too convincingly. Their concern is that AI-generated personas, such as chatbots with emotions, voices, and simulated personalities, may mislead users or create a psychological dependence.

Platforms are now expected to design AI in ways that "avoid confusion or misunderstanding," especially as conversational agents become more lifelike. Officials are wary of younger users trusting systems that are ultimately statistical constructs.

xAI Faces Another Legal Battle Over Data Center Pollution

Civil rights groups and environmental advocates are putting the spotlight on the infrastructure behind AI, alleging that the Elon Musk-owned xAI built and operated an unpermitted power plant to support its ever-growing data center footprint.

At least 27 gas turbines were reportedly installed at a facility in Southaven, Mississippi -- effectively creating a 495 MW power source to run its "Colossus II" data center. According to the NAACP's filing, the machines were deployed without the permits required by the Clean Air Act, despite emitting pollutants such as nitrogen oxides, particulate matter, and formaldehyde.

Community groups argue that the costs of this infrastructure, which includes pollution, health risks, and regulatory shortcuts, are being externalized onto local populations. If software once "ate the world," AI may now be consuming something more finite: energy.

Image Credit: Anthropic via X