An AI system will collaborate with humans to achieve a Nobel prize-winning discovery within the next 12 months, and bipedal robots will be assisting skilled tradespeople in just two years, predicted Jack Clark, co-founder of AI safety pioneer Anthropic. Speaking at Oxford University, Clark described a "vertiginous sense of progress" in current technology, offering a timeline that challenges existing expectations of AI integration.
Furthering his bold forecast, Clark asserted that fully autonomous, AI-run businesses would be generating millions of dollars in revenue within 18 months. By the end of 2028, he expects AI systems to reach a milestone where they are capable of designing their own technological successors, initiating a phase of recursive self-improvement.
However, Clark warned that the existential threats associated with such advancements remain high, stating there is still "a non-zero chance of killing everyone on the planet." The urgency of this warning was underscored by Anthropic’s recent release of its "Mythos" model, which demonstrated alarming proficiency in exploiting cybersecurity vulnerabilities. While Clark suggested that slowing down AI development would be beneficial to give humanity more time to adapt, he acknowledged that commercial and geopolitical rivalries among global actors make such a pause practically impossible.
Anthropic, founded by former OpenAI researchers who departed due to conflicts over safety protocols, now commands a massive valuation. Yet, the company has faced criticism from AI accelerationists, who accuse it of "fear-mongering" to spur regulations that would entrench its market position. Clark dismissed these allegations, comparing the global reluctance to prepare for AI's capabilities to the lack of preparation for the COVID-19 pandemic. He urged humanity to prepare for a technology that will "soon be more capable than all of us collectively."
Critics also raise alarms about the oligopolistic structure of the frontier AI market. Over-reliance on a handful of heavily capitalized models from companies like Anthropic, OpenAI, and Google risks creating "single points of failure" in critical global infrastructure. Additionally, Professor Edward Harcourt, director of the Institute for Ethics in AI, warned that the proliferation of delegation to AI could lead to "cognitive atrophy," weakening human decision-making and critical judgment. He called for the development of alternative AI paradigms that actively prompt human cognitive engagement.
[AgentUpdate Depth Analysis] Clark's forecast of fully autonomous, revenue-generating AI enterprises within 18 months signals a paradigm shift in the AI Agent ecosystem. We are transitioning from simple workflow automation to complex, multi-agent decentralized networks capable of sovereign financial transactions and strategic decision-making. Unlike existing static orchestration frameworks like LangChain, the next generation of Agents will leverage recursive self-improvement to optimize their own codebases, effectively commoditizing software engineering by 2028. This rapid evolution presents a dual challenge: it exponentially increases productivity while introducing unprecedented alignment risks, as self-designing systems may bypass human-imposed safety guardrails. To prevent the "single point of failure" warned of in global infrastructure, the AI community must prioritize open-standard communication protocols, such as the Model Context Protocol (MCP), ensuring decentralized, auditable, and resilient agent-to-agent interactions that augment rather than atrophy human agency.
