On May 7, 2026, Anthropic published the research agenda for The Anthropic Institute and said it is already seeing early signs of AI contributing to faster AI research and development. In Axios coverage published later, Anthropic co-founder Jack Clark said he now sees a greater than 60% chance that by the end of 2028 an AI system could autonomously build a better version of itself. That combination matters because it moves “intelligence explosion” talk from speculative safety debate into an official frontier-lab agenda with direct implications for enterprise AI governance.
What Anthropic published on May 7
Anthropic’s new agenda for The Anthropic Institute focuses on four research areas: economic diffusion, threats and resilience, AI systems in the wild, and AI-driven R&D. The most important line for business readers is that Anthropic says it is seeing early signs of AI helping speed up the research and development of AI itself.
The company also committed to publishing more granular labor-impact data through the Anthropic Economic Index, more research on resilience against AI-enabled security risks, and more detailed information about how Anthropic’s own work is speeding up because of new AI tools. Anthropic says this research should inform not only public debate but also its own release decisions and the work of its Long-Term Benefit Trust.
Why Jack Clark’s 2028 prediction is bigger than a provocative quote
Jack Clark’s estimate is not that fully autonomous self-improving AI has arrived in May 2026. The stronger point is that Anthropic now believes the path toward automated AI R&D is plausible enough to justify early-warning systems, governance exercises, and explicit public discussion. In Clark’s May 4 essay, he argued that frontier models are already strong enough at coding, tool use, and multi-step delegation to automate large parts of AI engineering, even if they are not yet reliably generating major scientific breakthroughs on their own.
That helps explain why Anthropic’s institute agenda includes questions about telemetry for AI R&D, intervention points for slowing acceleration, and even “fire drill” scenarios for an intelligence explosion. Labs do not usually build fire drills for scenarios they think are comfortably far away.
Business impact for enterprise AI teams
For most companies, the immediate takeaway is not to panic about recursive self-improvement inside their own stack. The practical takeaway is that enterprise AI programs should assume model capability jumps, agent autonomy, and safety expectations may keep arriving faster than internal controls, procurement processes, and review committees can adapt.
That changes three planning assumptions.
- Evaluation has to become continuous. If models keep gaining longer-horizon autonomy, point-in-time testing will age badly.
- Governance has to cover agent behavior, not just model access. The question is increasingly what systems can do across tools, approvals, and handoffs.
- Operational bottlenecks matter more than demos. If capability improves faster than workflows, auditability, and human override, the limiting factor becomes execution discipline rather than model quality alone.
This is especially relevant for companies building AI agents for security operations, customer support, internal knowledge work, software engineering, and back-office automation. Those are the categories where longer independent work loops and cross-tool action matter most.
What to watch next
The next signal is not whether Anthropic or another lab makes a dramatic AGI claim. It is whether frontier labs start publishing more hard telemetry on how much AI is speeding their internal R&D, whether governments begin formal acceleration-response planning, and whether enterprise buyers start demanding stronger evidence around agent controls before expanding deployments.
Anthropic has now put three ideas into the same public frame: labor disruption, cyber resilience, and AI-driven AI development. That is a notable shift. For Nerova readers, the implication is straightforward: the companies that benefit most from agentic AI over the next two years are likely to be the ones that pair automation ambition with stronger rollout discipline, clearer approvals, and tighter governance from the start.