Anthropic’s Secret Weapon Is Its Cult of Safety

(Bloomberg Opinion) — Silicon Valley’s most ideologically driven company may have become its most commercially dangerous.

This week’s $300 billion sell-off in software and financial services stocks was apparently triggered by a new legal product launched by Anthropic PBC and artificial intelligence startup. While you may think it’s pointless to attribute market debacles to a single trigger, concern over Anthropic’s disruption sheds light on its seemingly unstoppable productivity.

The company, which has about 2,000 employees, says it launched more than 30 products and features in January alone. On Thursday, it continued its momentum with Claude Opus 4.6, a new model designed to handle knowledge work tasks that will almost certainly increase tensions against legacy software-as-a-service (SaaS) companies like Salesforce and ServiceNow.

OpenAI, which owns ChatGPT, has a workforce twice the size of Anthropic, while Microsoft Corp. and Alphabet Inc.’s Google employ 228,000 and 183,000 staff respectively and have massive capital positions and distribution networks. But Anthropic’s AI tools for creating computer code and running computers go beyond anything these large companies have managed to bring to market. OpenAI and Microsoft have struggled to deliver such effective products lately.

Anthropic’s relentless efficiency comes in part from a paradoxical source: a mission-obsessed culture. It was founded by former staff at OpenAI who believed the company was too blasé about security, especially the existential risk AI poses to humanity. The topic has become something of an ideology at Anthropic, with Chief Priest and visionary Chief Executive Officer Dario Amodei.

Twice a month, Amodei brings his staff together for a Dario Vision Quest, or DVQ; Here, the bespectacled CEO will speak at length about creating trustworthy AI systems aligned with human values, broader topics like geopolitics, and the impact of Anthropic technology on the labor market. Amodei warned last May that AI advances could eliminate 50% of entry-level office jobs in the next one to five years; It’s an outcome his own company seems willing to fuel through its almost religious zeal for safe AI. Job security doesn’t seem to affect Anthropic’s ideals.

But people close to the company describe a cult-like atmosphere in which staff are focused on the mission and profess to believe in Amodei. “Ask anyone why they’re here,” Boris Cherny, one of the company’s chief engineers, told me recently. “Take them aside and they’ll tell you it’s about making AI safe. We exist to make AI safe.”

When leaders of Meta Platforms Inc. embarked on an expensive hiring spree for senior AI researchers last year, their approach to targeting Anthropic employees was to reassure them that Meta would move away from building open-source AI systems that are free to use and modify. Anthropic personnel found this approach fraught with danger.

Earlier this month, Amodei published a 20,000-word article about the risk to civilization recently posed by AI, while the company published a lengthy ‘constitution’ for its flagship system, Claude, suggesting that AI could have some form of consciousness or moral status.

While this is philosophically problematic, it is also a security measure; because the constitution intended to guide Claude gives clearer instructions on how to handle the possibility of being locked into the system (what he might see as death).

The company’s relentless focus on security has made its models some of the most honest on the market, according to a ranking by Scale.ai researchers powered by Meta; This means they are less likely to hallucinate and instead more likely to admit they don’t know something. This has made it more reliable for a growing number of corporate customers.

This obsession, too, is unusual in an industry prone to mission drift, where tech companies are built on noble ideas to improve humanity before inheriting obligations to investors. Remember Google’s “don’t be evil” slogan? OpenAI, which was founded to “benefit humanity” as a non-profit organization unhindered by financial constraints, is another example of this.

But Anthropic’s mission-driven culture has the added benefit of eliminating the kind of internal friction that slows things down in the corporate bureaucracies of Google and Microsoft; because the staff works harmoniously to achieve the company’s goal. As a result, security-obsessed Amodei, who has the uneasy look of a mad scientist, is shipping more products than some of the biggest names in Silicon Valley.

“Military historians often argue that the sense of fighting for a noble cause drives militaries to perform better,” says Sebastian Mallaby, author of a forthcoming book on Google DeepMind and a senior fellow in international economics at the Council on Foreign Relations. He says Anthropic’s advantage lies in the fact that, unlike OpenAI, which pursues multiple paths, it focuses on building a highly effective coding tool known as Claude Code and gaining enterprise customers on that basis. He adds that Sam Altman’s company suffers from “the arrogance of leaders.”

Anthropic is currently raising $10 billion at a $350 billion valuation; This means the pressure to prioritize growth over security will become even more intense. Amodei built a sending culture. The question is whether this culture can continue when the risks increase and continue to shake markets and potentially many businesses. More from Bloomberg Opinion:

This column reflects the author’s personal views and do not necessarily reflect the views of the editorial board or Bloomberg LP and its owners.

Parmy Olson is a Bloomberg Opinion columnist who covers technology. The former reporter for the Wall Street Journal and Forbes is the author of “Supremacy: AI, ChatGPT, and the Race to Change the World.”

More stories like this available Bloomberg.com/opinion