Could AI really wipe out humanity and hijack the entire internet? That is the alarming question raised by recent claims from AI leaders, including Anthropic CEO Dario Amodei, who warned that within 6–12 months a swarm of AIs could take over the internet with a persistent botnet causing hundreds of billions in damage. We examine six of the most shocking AI safety claims and separate hype from genuine risk.
The Six Claims Driving the AI Safety Debate
The debate exploded after a series of statements from researchers, executives, and critics. Some warn of a 10% chance of human extinction; others call AIs worse than nuclear weapons. A third group insists the entire internet is vulnerable to an AI-driven botnet, while sceptics dismiss it all as a big tech psyop. Here is what each claim actually says.
Claim 1: A 10% Chance of Doom
Several prominent AI researchers have assigned a roughly 10% probability to catastrophic outcomes from advanced AI. While speculative, the figure is not pulled from thin air: it reflects deep uncertainty about alignment and control. Treat it as a risk warning, not a prophecy.
Claim 2: AIs Are Worse Than Nukes
Comparing AI to nuclear weapons is rhetorically powerful but technically loose. Nukes are physical, verifiable, and state-controlled. AI is digital, copyable, and diffuse. The comparison highlights stakes, not mechanisms.
Claim 3: The Internet-Hijacking Botnet
Dario Amodei’s botnet warning is the most concrete. Yet AI sceptic Gary Marcus called the idea that the entire internet could be taken over “nonsensical.” He points to a UK AI Security Institute assessment finding that Mythos, one of Anthropic’s leading AIs, could only “autonomously compromise small, weakly defended vulnerable systems.”
Claim 4: It’s All a Big Tech Psyop
Some critics argue AI doom talk is a marketing tactic to attract regulation that locks in incumbents. There is a kernel of truth: fear drives funding and policy. But dismissing all risk as a psyop ignores real vulnerabilities in poorly defended systems.
Claim 5: AI Will Autonomously Hack Critical Infrastructure
Large parts of the internet run on infrastructure from Cloudflare, Google, and AWS. Marcus and colleagues doubt any of these could be taken down or controlled for long without government force. Individual sites may be attacked, but the core is resilient.
Claim 6: Regulation Will Save Us
Regulation can help, but it lags technology. The more practical near-term defence is basic security hygiene: patching, segmentation, and monitoring.
Comparing the Claims: Risk vs. Reality
| Claim | Proponent | Plausibility | Key Caveat |
|---|---|---|---|
| 10% chance of doom | Various researchers | Low–moderate | Highly speculative |
| AI worse than nukes | AI safety advocates | Low | Different risk class |
| Internet-wide botnet | Dario Amodei | Very low | Requires extreme negligence |
| Big tech psyop | Sceptics | Partially true | Ignores real flaws |
| Autonomous critical infrastructure hack | Some experts | Low near-term | Core providers are hardened |
| Regulation as saviour | Policymakers | Moderate | Slow and uneven |
Key Takeaways
- AI doom claims are overstated but not entirely baseless.
- Full internet takeover is implausible given hardened core infrastructure.
- Weakly defended systems remain at risk from autonomous AI attacks.
- Scepticism is healthy, but so is basic cybersecurity hygiene.
- Regulation alone won’t save you; proactive defence matters.
What This Means for Businesses and Users
For most organisations, the practical threat is not a global AI takeover but targeted attacks on unpatched systems. Ensure backups, multi-factor authentication, and network segmentation. For individuals, the risk is mostly indirect: phishing, fraud, and data breaches enabled by AI-generated content.
Rather than panic, treat AI safety claims as a prompt to audit your own digital resilience. The internet’s core is stronger than headlines suggest, but the edges are where breaches happen.
FAQ
Could AI really take over the entire internet?
Highly unlikely. Core infrastructure from Cloudflare, Google, and AWS is heavily defended. AI could compromise small, weakly defended systems, but a full internet takeover would require extreme negligence.
Is the 10% chance of AI doom credible?
It is a speculative estimate, not a scientific consensus. It reflects deep uncertainty about advanced AI alignment and should be treated as a risk warning rather than a prediction.
What should I do to protect myself from AI-driven cyberattacks?
Focus on basic security: patch systems, use multi-factor authentication, segment networks, and train staff against phishing. These measures defend against most realistic AI-enabled threats.
Best Products We’ve Tested and Rated

Our testing team has hands-on reviews of extendable outdoor dining tables, quick access gun safe, rifle, handgun car, and fireproof waterproof. Every option below was compared across price, build quality, and real-world performance, with honest pros and cons. We update these guides regularly as new models arrive, so the recommendations stay current.
Our testing team has hands-on reviews of gun money, home, home centric smart, thermal camera what it is how choose model, and how buy armored door home. Every option below was compared across price, build quality, and real-world performance, with honest pros and cons. We update these guides regularly as new models arrive, so the recommendations stay current.