Anthropic Co-Creates a Warning: 50% Chance the Software Singularity Will Arrive—Could It Hit Both White-Collar and Blue-Collar Workers?

Anthropic senior executives are visiting Taiwan, emphasizing that without Taiwan, there would be no development behind its nearly $1 trillion valuation. The co-founders also warned that there is a 50% chance of a software singularity in the future, and that knowledge workers would be the first to be hit by unemployment.

During Computex, the US AI company Anthropic—whose valuation is edging close to $1 trillion—was also moving forward along the road to an IPO, deepening its cooperation with Taiwan.

Next came the news that Trend Micro was invited to join the cybersecurity program “Glasswing.” At a June 5 event, Anthropic co-founder Ben Mann called out, “Without Taiwan, there is no Anthropic,” and delivered a complete opening speech in Chinese, showing his approachability.

At a Developer Day hosted by the Taiwan AI School (AIA), Mann said in Chinese that his mission in undertaking AI research and development is to ensure that artificial intelligence can have a positive impact on humanity’s future. He believes strongly that democratic countries should take the lead in AI development and looks forward to deepening cooperation with Taiwan.

Before founding Anthropic in 2021 with Dario Amodei, Ben Mann previously spent six years at Google. After reading the book “Superintelligence,” he shifted his focus to AI safety research.

Mann was previously a core member of OpenAI’s GPT-3 core architecture and is an important driving force behind MCP (Model Context Protocol). He currently leads “Anthropic Labs,” the department within Anthropic that focuses on applied experiments—this team incubated Claude Code and Claude Design.

AI toward a software singularity—mass unemployment is unavoidable?

When talking about the capabilities of Mythos-level models, Mann pointed out that a sudden leap in programming ability will enable more people to safely build software applications on their own. “In the past, software engineers who did this kind of work—where do they go from there? I think this is a bigger problem.”

Rather than whether AI can think like humans, Mann is more focused on the real extent of the transformation AI will bring on the economic and social fronts. He has proposed the concept of an “Economic Turing Test” as a tool for measuring economic output: if work is outsourced to an AI agent and the employer cannot tell whether the results come from a machine or a human, then it passes that job’s Economic Turing Test.

Image source: Digital Times, photo by Hou Junwei Anthropic co-founder and head of Anthropic Labs Ben Mann visiting Taiwan

Mann is quite confident, stating that within the next few years there is a 50% chance of a software singularity in the software field. At that point, the problem of controlling robots could also be solved. As a result, first knowledge workers, and then physical laborers, will all face disruptions in the job market.

Mann said that current economic indicator data does not suggest that large-scale unemployment is about to happen, “but this actually matches our expectations.” He explained that to drive large-scale unemployment, AI must be able to do every thing humans can do—and AI has not yet reached that level.

However, there is no sign that progress in model training is slowing down. Mann gave an example: just as most people couldn’t predict the 2008 financial crisis by observing indicators, “I believe we’re going to go through a bumpy ride.”

Mann’s view is not a consensus in the industry. Nvidia CEO Jensen Huang has generally been more optimistic about AI’s impact, believing it could create more job opportunities. Anthropic CEO Amodei, who previously called for a “white-collar slaughter” in 2025, has recently also softened his public stance slightly, pointing to a shift in jobs rather than their disappearance.

Mann believes that the problem of large-scale unemployment is difficult to solve without government involvement. Anthropic is also working with governments and discussing these issues; he hopes Taiwan can “move faster than some governments around the world.”

A Biden administration tech and national security policymaker visits Taiwan together—team up to ensure the democratic alliance leads in the AI race

Joining Mann on his visit to Taiwan is Tarun Chhabra, Anthropic’s head of national security policy. He was a core figure in the Biden administration’s tech and national security policies, where behind-the-scenes decisions cover everything from the US’s export controls toward China to investment screening for entry and exit, and restrictions on information and communications technology.

During his time in the White House, building a democratic technology alliance was another priority for Chhabra. He conducted technology diplomacy with the Five Eyes, Japan, South Korea, India, and the European Union, aiming to bring international allies into the US-led defensive line of technology standards.

Image source: Digital Times, photo by Hou Junwei Anthropic’s head of national security policy Tarun Chhabra talks with Taiwan’s Deputy Minister of Digital Development Yang Jialing and Professor Xu Yongzhen from the Taiwan AICoE project office.

This context makes the strategic significance of Anthropic’s senior-level visit to Taiwan far beyond simply exchanging ideas with developers—it underscores a clear Taiwan-US partnership in technology and supply-chain collaboration.

During the discussion, Mann said plainly that the democratic alliance should prevent China from winning the AI race. Even if the democratic camp gets ahead and achieves superintelligence first, vigilance cannot be relaxed, because other countries’ R&D will keep progressing. He said, “Anthropic has always tried to use our leading position to set a strong, powerful precedent for improving safety.”

He gave an example: before releasing a new model, Anthropic tested whether the model’s capabilities could accelerate the development of biological weapons. The result was that one sample showed a significant improvement. The company therefore delayed the model release and only released it after strengthening protective measures.

This policy approach can also be seen in the latest Mythos model. Mythos’s ability to find software security vulnerabilities exceeds most humans’ capabilities. As a result, Anthropic launched “Project Glasswing,” releasing only a preview version to a small number of verified partner companies, in order to strengthen security defenses in advance.

Mann said the company understands that delaying the release of cutting-edge models can reduce the risk of malicious misuse, but it also delays the real-world delivery of benefits that the models could bring. This includes enabling more cybersecurity defenders to strengthen their own security defenses and accelerating drug development to save patients with serious illnesses. “We truly feel the cost of delaying releases.” However, safety remains the prerequisite.

“At Anthropic, we are deeply thinking about these risks,” Mann said. “As technology becomes more and more powerful, how do we ensure that countries that do not share the same values as Taiwan use this technology responsibly?”

If superintelligence is so dangerous, why develop it? In the discussion, Chhabra provided an answer: “If Anthropic doesn’t develop this model, our competitors will.” To prevent superintelligence from being misused first by malicious actors, Anthropic must be ahead in setting safety standards. “We’re racing against time.”

Echoing Mann’s remarks, Chhabra reiterated in the discussion that without Taiwan’s supply chain, there would be no Anthropic. For democratic countries to win the AI competition, “jointly ensuring the security of this supply chain is absolutely critical and necessary.”

Chhabra also pointed out that Taiwan is often the first place to face advanced cyberattacks, and that such attack techniques are highly valuable information for allies. He hopes that through partnerships, Taiwan and the US can demonstrate strong resilience and a solid supply chain to all democratic allies.

• This article is reprinted with permission from: Digital Times
• Original title: Anthropic co-founder warns in Taiwan: “Software singularity” has a 50% chance of arriving, with knowledge workers hit first
• Original author: Fu Wanqi