A Story of Two Anthropics

(SeaPRwire) –   On Wednesday, I was at an East London event space filled with developers. The occasion was Code w/ Claude: Anthropic’s first Europe-based gathering focused on developers. It felt like a celebratory tour for a company reportedly raising funds at a $900 billion valuation—if accurate, that number would surpass key rival OpenAI. Behind Anthropic’s rapid rise are the developers who packed the London venue this week, enjoying free meals and taking home complimentary mini computers. These power users spend far more on Claude than the average person. In short, they are Anthropic’s lifeblood, and the company has every reason to keep them coming back.

Throughout the day’s keynotes across two stages, Anthropic executives poured out a flood of technical details on how coders can best use AI to boost productivity. The room’s mood was one of nearly unbridled enthusiasm. Boris Cherny, creator of Claude Code, shared that AI had reconnected him with the sense of magic that first drew him to programming on a school calculator. Developers I spoke to talked of a renaissance in computer programming, with more opportunities available than people to seize them.

The day painted Anthropic as a product-focused company. There was little mention of the risks Anthropic has been vocal about elsewhere. On stage, staff alluded to ongoing AI progress—using many upward-trending graphs—but mostly to encourage developers to harness Claude Code’s capabilities, not to mitigate risks. When an Anthropic researcher asked how many had shipped Claude-written code without reading it, a surprising number raised their hands. “That’s a dangerous game,” he said, to nervous laughter, then quickly shifted to how developers could maximize Claude’s potential.

Thursday was a starkly different experience. I took a train to Oxford University, where Anthropic co-founder Jack Clark was set to lecture. Clark is a fascinating figure; much of his work has involved measuring and warning about AI progress and its attendant risks. This lecture was no exception: he said AI carries a “non-zero” chance of killing everyone on Earth, and even if that crisis is averted, the next few years will bring more disruption than any in living memory.

Clark predicted AI would reach “recursive self-improvement” (the ability to improve itself) by 2028 or sooner, leading to an unpredictable future. “We’ve never encountered a technology that can make a better version of itself without human input,” he said. Most of the world is “in denial” about current AI capabilities, let alone those arriving in six months, he added.

It would be easy—and comforting—to dismiss this as a billionaire hyping his tech for self-interest. But another interpretation is that this lecture (and much of Anthropic’s work) is a warning from those closest to this strange new technology, who see its path. The April launch of Claude Mythos—Anthropic’s latest model with nation-state-level cyber-offensive capabilities—proves many past predictions were correct, and the future they foresaw is already here in some ways. The model is so powerful Anthropic chose not to release it publicly, instead giving a small group of companies and governments access to patch vulnerabilities in the world’s most used software.

Clark said even Anthropic underestimated AI’s scale and speed. When Mythos finished training, he recalled: “We thought, ‘Oh, it’s here! Faster than we expected, and we haven’t prepared enough.’”

It’s neither nefarious nor surprising that Anthropic speaks differently to policymakers than developers—companies tailor messages to audiences all the time, and Anthropic has been open about AI risks where others stay silent. Still, experiencing these two narratives back-to-back gave me a strong sense of whiplash. On one hand, Anthropic plausibly holds the keys to the most powerful tech ever, while warning societies of that inconvenient fact. On the other, it’s still a normal software company, dependent on the whims of casual developers in hoodies and jeans.

If Clark is right, this moment won’t last long. The developers Anthropic now relies on will be replaced by machines faster than they can say “recursive self-improvement.” Maybe that’s why his predictions didn’t make it into Wednesday’s keynotes.