Artificial intelligence startup Anthropic is making its most advanced AI technology available to a broader audience for the first time, while imposing strict safeguards designed to prevent the system from being used for potentially dangerous cybersecurity activities.

The company on Tuesday unveiled Claude Fable 5, describing it as the most capable model it has released for general use. The launch marks a significant milestone for Anthropic as it seeks to expand commercial adoption of its technology following the global attention generated by its experimental Mythos model earlier this year.

Mythos stunned governments, cybersecurity experts, and technology companies after Anthropic disclosed that the system had identified thousands of previously unknown software vulnerabilities, raising both excitement about its capabilities and concern about how such powerful tools could be misused.

Until now, access to Mythos has been restricted largely to about 200 organizations participating in Anthropic’s Glasswing program, including U.S. government agencies and selected cybersecurity institutions.

The wider rollout comes at a crucial moment for Anthropic. The company has emerged as one of the most valuable businesses in the artificial intelligence sector, carrying an estimated valuation of $965 billion and increasingly competing head-to-head with rivals such as OpenAI, Google DeepMind, and xAI as the race toward public listings accelerates.

Balancing Power and Safety

Anthropic’s challenge has been how to commercialize a model capable of sophisticated cybersecurity reasoning without enabling malicious actors to exploit those same capabilities. The company said Claude Fable 5 includes extensive safeguards that block users from employing the model to search for software vulnerabilities, conduct offensive cyber operations, or perform other high-risk activities.

Dianne Penn, Anthropic’s head of product management, research, and labs, explained how the restrictions would work in practice.

“Let’s say I’m a college student asking the model like help me find cyber vulnerabilities on X package or code. The model would refuse and Fable 5 will fall back to Opus 4.8 for a response,” Penn told Reuters.

The approach reflects a shift across the AI industry, with companies separating frontier research systems from commercially available products. Rather than releasing raw capabilities directly to consumers, firms are embedding policy controls that limit access to sensitive functions.

Anthropic said it conducted extensive testing to ensure users cannot easily manipulate or jailbreak the system into performing restricted actions.

The original Mythos preview generated attention because it demonstrated capabilities that went beyond traditional coding assistants.

According to Anthropic, the model was able to uncover large numbers of software vulnerabilities, highlighting how advanced AI systems may soon become powerful tools for both cybersecurity defense and cyber offense.

That dual-use nature has become one of the most contentious issues in artificial intelligence development.

Governments view advanced AI models as potential national security assets. The U.S. government has expanded cooperation with Anthropic through initiatives such as Glasswing, while countries including South Korea, Japan, and several European allies have sought access to frontier AI systems for cybersecurity purposes.

The debate has intensified as researchers warn that future generations of AI could autonomously discover and exploit vulnerabilities faster than human experts. Anthropic’s decision to keep unrestricted Mythos access confined to trusted organizations appears aimed at reducing those risks while still enabling legitimate research and security applications.

Anthropic recently confidentially filed for an initial public offering, placing it among a growing list of AI giants preparing to enter public markets. Investors are focused on whether frontier AI developers can convert technological leadership into sustainable revenue growth.

By releasing Claude Fable 5 more broadly, Anthropic gains an opportunity to expand enterprise adoption while preserving safeguards around its most sensitive capabilities. The company said customers already using the preview version of Mythos will be able to upgrade to Claude Mythos 5, while broader access will gradually expand through what it described as a more systematic trusted-access program.

This tiered approach allows Anthropic to monetize cutting-edge technology without fully relinquishing control over its most powerful capabilities.

Cost Efficiency Becomes a New Battleground

Beyond performance, Anthropic is also emphasizing efficiency. Penn said early users have reported that while Fable 5 is positioned as a premium model, it completes tasks using fewer tokens, potentially lowering overall costs for customers.

As enterprises spend billions of dollars deploying AI systems, attention is increasingly moving away from raw model capability toward economics. Companies are now evaluating not only which models perform best, but also which deliver the lowest cost per completed task.

Competition on efficiency is becoming important as businesses scrutinize soaring AI budgets and seek ways to manage spending without sacrificing performance. Anthropic has priced both Claude Fable 5 and Claude Mythos 5 at $10 per million input tokens and $50 per million output tokens.

The artificial intelligence industry’s most advanced models are becoming powerful enough to influence national security, software infrastructure, and critical systems. Yet the same capabilities that make them valuable also create concerns about misuse. Anthropic’s solution is effectively a compromise: make the technology more widely available while limiting access to its most sensitive cybersecurity functions.

Fears that artificial intelligence is already wiping out white-collar jobs may be running ahead of the evidence, according to a senior economist at Google DeepMind, who says the data so far does not support claims of a widespread employment crisis.

Alex Imas, director of AGI economics at Google DeepMind and a professor at the University of Chicago, said he has yet to see convincing signs that AI is causing broad-based layoffs across professional occupations, even in sectors considered most exposed to automation.

Speaking on the Dwarkesh Podcast, Imas pushed back against growing predictions from some technology leaders that AI will soon eliminate large numbers of office jobs.

“A lot of people are looking at it,” he said, adding that “even looking at software engineering, the most exposed sectors, there’s just not really anything going on.”

This stands in contrast to dire warnings from some leaders at frontier AI companies. Executives such as Dario Amodei have cautioned that AI could eventually remove large portions of entry-level white-collar work, while other industry figures have warned of significant labor market disruption over the next several years.

Imas, however, argued that current labor market data does not show evidence of a large-scale employment shock.

“Right now, we don’t really have any evidence of a white-collar bloodbath,” he said.

The Bigger Risk May Be Corporate Psychology

While dismissing claims of an ongoing jobs apocalypse, Imas highlighted a different risk that could emerge if corporate behavior becomes driven by perceptions rather than economic reality. He described a hypothetical scenario in which companies begin cutting staff simply because investors and competitors expect AI adoption to translate into workforce reductions.

“Let’s say we get into a narrative where if you’re a firm and you’re not laying people off, then you’re seen as not adapting AI enough,” he said.

“That’s super worrying, where the firm might actually be worse off after the layoffs than before the layoffs, but it’s just doing the layoffs to have the perception that, ‘Look, we’re not behind the times. We’re using AI.'”

Such a dynamic could create what economists often describe as a herd effect, where businesses mimic one another’s actions not because the strategy improves productivity, but because they fear appearing less innovative than rivals.

The warning comes as corporate executives face mounting pressure from investors to demonstrate clear AI strategies. In recent months, several technology firms have cited AI as a factor in workforce reductions, fueling speculation that automation-driven restructuring is accelerating.

Rather than eliminating jobs outright, Imas suggested AI’s immediate impact may be to automate portions of existing roles while leaving workers responsible for higher-value activities.

“Let’s say the AI automates nine out of ten tasks. One task is not automated,” he said.

“If that person can now focus in on that task, the job will become more productive.”

This view aligns with a growing body of economic research suggesting that AI’s near-term effect may be augmentation rather than wholesale replacement. Historically, major technological advances have often increased productivity by changing how work is performed rather than eliminating occupations entirely.

That does not mean labor markets will remain unchanged. Certain routine tasks may disappear, new roles may emerge, and skill requirements could shift significantly. But the transition may be more gradual than some forecasts suggest.

A Debate Dividing the AI Industry

A widening divide within the AI sector itself has been going on. On one side are executives who warn that capable AI systems could rapidly replace knowledge workers across industries. On the other hand, there are researchers and economists who argue that there is currently little empirical evidence supporting predictions of mass unemployment.

Notably, a spokesperson for Google DeepMind emphasized that Imas was speaking in a personal capacity and that the scenario he described was hypothetical. The spokesperson added that his broader point was that existing data does not indicate a white-collar employment collapse.

The company also pointed to previous remarks by Demis Hassabis, who has argued that AI could ultimately boost productivity, create new industries, and generate entirely new categories of jobs. The debate is becoming increasingly important as businesses invest billions of dollars into AI systems and governments assess how the technology may reshape labor markets.

For now, according to Imas, the evidence suggests caution against dramatic conclusions.

The technology is advancing rapidly, but the widely predicted wave of AI-driven white-collar job destruction has yet to materialize in the data. The greater immediate risk, he suggests, may be companies acting on the expectation of disruption before the disruption itself has actually arrived.