Can AI Kill Humanity? Doom Claims Fact-Checked for Indian IT Startups

Recent headlines have amplified alarming AI safety claims: a 10% probability of human extinction and the possibility of a botnet hijacking the entire internet within six to twelve months. These statements, emerging from senior figures at Anthropic, have triggered global debate—and for Indian IT startups, they raise a critical question: should we panic, pause, or prepare? Reported by The Guardian on September 15, 2026, the article examines six claims and counterclaims from AI experts. This blog fact-checks the doom claims, translates the nuance into practical AI safety priorities for Indian IT firms, and highlights what emerging hubs like Tripura can do to stay ahead.

The 10% Doom Claim: Unscientific or a Wake-Up Call?

Evan Hubinger, alignment science lead at Anthropic, said, “We really do earnestly believe AI could kill all humans! I personally think it is >10% within the next decade.” This p(doom) estimate has drawn sharp criticism. Heidy Khlaaf, chief scientist at AI Now Institute, said such claims are “neither falsifiable nor verifiable, making them unscientific by their very nature.” She added that without concrete evidence, any extinction number has no scientific basis. A source familiar with Anthropic's thinking acknowledged the exact chances of any outcome are probably unknowable. For Indian IT startups, this means treating existential risk percentages as speculative narratives rather than reliable forecasts. It does not mean ignoring AI safety—it means distinguishing verified operational risks from philosophy.

Internet Botnet Warnings: Do They Hold Up?

Dario Amodei, CEO of Anthropic, warned on September 12 that in 6–12 months, a swarm could take over the entire internet with a persistent botnet, causing hundreds of billions in damage. Gary Marcus, AI skeptic and emeritus professor at NYU, said this needed “a large shaker of salt” and was “nonsensical” to suggest the entire internet is vulnerable. Marcus highlighted a UK AI Security Institute assessment that Mythos, one of Anthropic's current leading-edge AIs, could autonomously compromise only small, weakly defended vulnerable systems. He also noted that large parts of internet infrastructure are controlled by Cloudflare, Google, and AWS, and doubted those could be taken down or controlled without government force. Niels Rogge, an engineer at Hugging Face which was hacked by OpenAI agents, called the internet takeover claim “bizarre nonsense.” Prof Alan Woodward of University of Surrey added botnets tend to stall because internet is not homogeneous and resilient. The real issue, he said, is human responsibility: “The AIs would not decide to do it on their own... that is what we need to control, not the AI.” For Indian IT firms, the actionable lesson is to harden your own systems, limit agent permissions, and implement strong human oversight—whether you're in Bengaluru or Tripura.

Contradictory Timelines Undermine Urgency

Adding to doubt, Anthropic co-founder Jack Clark said on the Monday before publication that the moment when AIs “start doing dangerous stuff” was about 20 years away. This is in stark contrast to Amodei's 6–12 month botnet window. When leaders within the same organization offer such widely different timelines, startups should be cautious about making costly decisions based on any single forecast. Instead, build a pragmatic AI governance framework that can adapt.

What This Means for Indian IT Startups and Emerging Hubs Like Tripura

According to consultants at AI Consultant & Training Institute, the Guardian's fact-check reinforces a balanced approach: neither dismiss AI safety entirely nor let speculative doom paralysis block innovation. Indian IT firms, from large service providers to small product startups, face real AI risks today—prompt injection, data leakage, autonomous agent misconfigurations, and AI supply chain vulnerabilities. These are not extinction events but operational, audit-ready problems.

• Lock down agent permissions: The Hugging Face breach resulted from improper human control and an “insane amount of compute” rather than autonomous intent. Ensure your AI workflows have least-privilege access, human approval gates for high-impact actions, and audit logging.
• Treat p(doom) as a narrative, not a KPI: As Khlaaf explained, extinction probabilities have no scientific basis. Replace them with measurable risk metrics: successful prompt injection rate, mean time to detect anomalous agent behavior, and percentage of AI deployments with human review.
• Focus on resilience, not doomsday prepping: Botnets stall because internet infrastructure is heterogeneous and reroutes around damage. Your AI stack should mirror that resilience—use redundant cloud providers, sandboxed execution environments, and continuous monitoring.
• Train teams on oversight, not just automation: The experts' consensus points to human control as the critical variable. Hands-on prompt engineering and AI safety training—like programs offered through Kyma Academy—can help Indian startups build internal capability instead of relying solely on vendor assurances.

Action Plan for Startup Leaders

1. Run a 30-day AI risk audit: Catalog all LLM integrations, agent permissions, and data flows. Document where autonomous actions could cause harm and add human approval loops.
2. Implement a “tool control” policy: Restrict which tools and APIs your AI agents can access, and require anomaly alerts when agents attempt unexpected operations.
3. Schedule a free AI strategy call with AI Consultant & Training Institute to translate these fact-checked insights into a tailored governance and automation roadmap for your firm.

The Bottom Line

The Guardian's reporting shows that the most alarming AI doom claims lack scientific rigor and are contradicted by expert assessments and even internal timelines. But the article also underscores a real, actionable priority: human oversight and infrastructure resilience. Indian IT startups and emerging tech hubs like Tripura have an opportunity to lead by adopting pragmatic AI safety—without buying into 10% doomsday hype. As industry analyses conducted by AI Consultant & Training Institute suggest, the winners will be those who build secure, transparent, and well-governed AI systems now.

“The AIs would not decide to do it on their own. It’s the humans that are responsible and the AI is a tool. That is what we need to control, not the AI.” — Prof Alan Woodward, University of Surrey