# Anthropic Researchers Raise Extinction Fears Over Runaway AI, Say Control Could Be Lost Within Years

*Wednesday, September 9, 2026 at 4:11 PM UTC — Hamer Intelligence Services Desk*

**Published**: 2026-09-09T16:11:09.934Z (2h ago)
**Category**: intelligence | **Region**: Global
**Importance**: 8/10
**Sources**: OSINT
**Permalink**: https://hamerintel.com/data/articles/17369.md
**Source**: https://hamerintel.com/summaries

---

**Deck**: Three researchers at AI firm Anthropic have gone public with concerns that uncontrolled artificial intelligence could wipe out humanity within this decade, arguing that current safeguards and regulation lag far behind model capabilities, according to Axios.

Warnings from inside a leading AI lab are sharpening a question for governments: what if the gravest risks from artificial intelligence arrive much sooner than expected?

According to Axios, three researchers at Anthropic, a major AI company, have publicly voiced fears that advanced systems could lead to human extinction within this decade. They argue that current safety measures and regulatory efforts are far behind the pace at which powerful models are being developed and deployed.

One of the researchers, Jacob Coxon, wrote that there is a real risk of losing control of increasingly capable systems, with consequences they believe could be catastrophic for human life if development continues without far stronger safeguards.

Anthropic’s teams work directly on large models now being embedded into productivity tools, customer‑service bots, coding assistants and security‑related applications. Their concerns focus on what happens as systems that already plan tasks and write code are scaled up and granted more autonomy.

The scenario they worry about is not limited to science‑fiction images of robots, but to models that can design sophisticated cyberattacks, help discover and exploit vulnerabilities, generate persuasive disinformation at scale, or assist in the design of biological threats. Once such systems are deeply integrated into critical infrastructure and decision‑making, they argue, close monitoring and emergency shutdowns may become harder.

For policymakers who have so far concentrated on issues like data protection, transparency and bias, the Anthropic warnings push AI into the realm of hard security. Legislators in the U.S., EU and elsewhere are still defining baseline rules for current‑generation tools at the same time militaries and intelligence services explore how to use AI for targeting, surveillance and cyber defense.

There is also a strategic dilemma. States and companies see AI as a major source of economic and military advantage and are worried about falling behind rivals. The researchers’ message is that without strong guardrails, competitive pressure could drive the release of systems whose risks aren’t fully understood.

Signals to watch now are whether regulators move toward binding safety standards for the most capable models, whether major labs commit to slower deployment when internal risk indicators rise, and how national‑security agencies build catastrophic‑risk scenarios into their planning. Concrete steps such as independent stress‑testing, incident reporting and limits on certain autonomous uses will show whether the warnings are changing practice or just adding to the debate.