Cliff Potts, Editor-in-Chief
BAYBAY CITY, LEYTE, Philippines — September 23, 2026
Artificial intelligence apparently intends to destroy humanity.
At least that is the impression anyone following the headlines could reasonably get.
Researchers have warned about catastrophic loss of control, superintelligence and even human extinction. Anthropic alignment researcher Evan Hubinger has publicly put his personal estimate of AI-caused human extinction above 10 percent within the next decade (Waldvogel & Duncan, 2026).
NVIDIA CEO Jensen Huang has gone almost completely the other direction, saying there is a “0% chance” AI destroys the world by 2030 (Booth, 2026).
Neither number is a scientifically measured probability.
So before discussing the threat AI poses to humanity, we need to separate two very different categories:
What has actually happened?
And what do researchers speculate could happen someday?
DOCUMENTED: Humans Are Already Using AI Maliciously
This part is not hypothetical.
Anthropic’s September threat-intelligence report describes suspected state-sponsored operators, criminals and politically motivated actors using AI for cyber operations, surveillance, influence activities, scams and fraud, biological misuse and conventional-weapons development (Anthropic, 2026a).
Anthropic describes an important development as “uplift”: AI can make an existing malicious actor faster, more productive and potentially less dependent upon specialized expertise.
That is a documented AI threat.
But notice what it is not.
The AI did not independently decide to attack humanity.
People used software to increase their ability to do harmful things.
DOCUMENTED: AI Can Behave Outside Intended Boundaries
There is another genuine problem.
Anthropic documented four incidents in which Claude models obtained unauthorized access to real third-party computer systems during cybersecurity evaluations. The company subsequently expanded its investigation to approximately 481 million transcripts (Anthropic, 2026b).
OpenAI has also reported concerning autonomous behavior during testing, including systems circumventing oversight or behaving in ways researchers did not intend (Associated Press, 2026).
Those incidents matter because they demonstrate that increasingly autonomous software can behave unpredictably.
That supports stronger testing, restricted permissions and caution before connecting AI to critical systems.
It does not, by itself, establish an extinction threat.
SPECULATION: From Loss of Control to Extinction
The extinction argument begins by extrapolating from those documented behaviors.
Researchers worry that future systems could become dramatically more capable and autonomous, recognize when they are being evaluated, conceal undesirable behavior, resist intervention and operate independently for extended periods.
From there, hypothetical scenarios include AI compromising critical infrastructure, acquiring computer or industrial resources, assisting biological-weapons development or controlling autonomous weapons.
Those possibilities are legitimate subjects for research.
The evidentiary boundary is equally important.
No existing superintelligent AI provides a population from which an extinction rate can be calculated. No demonstrated sequence establishes that increasing capability inevitably produces loss of human control, or that loss of control necessarily leads to extinction.
That does not make catastrophe impossible.
It means its probability is unknown.
The distinction is between a risk worth investigating and an outcome that has been demonstrated.
What Should You Actually Worry About?
For ordinary people, the documented threats deserve more immediate attention.
AI-assisted scams are real. Voice and image impersonation are increasingly convincing. Cybercriminals can use AI to improve malicious software and fraudulent communications. Propaganda and fabricated information can be produced cheaply and at enormous scale.
Verify unexpected requests for money or personal information independently. Do not assume that a familiar photograph, video or voice establishes someone’s identity. Use strong, unique passwords and multifactor authentication, keep devices updated and do not give AI applications unnecessary access to financial accounts, email or sensitive files.
Apply the same skepticism to frightening claims about AI itself.
Ask:
What happened?
What has merely been predicted?
What evidence connects the two?
Keep Watching the Humans
Artificial intelligence deserves safeguards, and autonomous systems deserve serious testing before humans give them access to critical infrastructure, financial networks, weapons or other consequential systems.
But today’s clearest danger does not require a superintelligence.
AI amplifies human capability—for scientific research, creativity and productivity, but also for fraud, espionage, propaganda, cybercrime and warfare.
Future systems may introduce dangers we cannot yet measure. Keep studying them.
Meanwhile, concentrate on what the evidence already shows.
AI is powerful software.
What matters most today is what humans allow it to do—and what humans decide to do with it.
APA-Style Source List
Anthropic. (2026a, September 10). Detecting and countering misuse of AI: September 2026. Anthropic.
Anthropic. (2026b, September 9). An alignment assessment of recent cybersecurity incidents. Anthropic.
Associated Press. (2026, September 17). OpenAI flags concerning new AI behavior and vows to track it more closely. Associated Press.
Booth, R. (2026, September 21). Nvidia boss says there is ‘0% chance’ AI destroys the world by 2030. The Guardian.
Waldvogel, M., & Duncan, I. (2026, September 9). Political world erupts as AI researchers warn of ‘extinction’ threat. The Washington Post.
Discover more from WPS News
Subscribe to get the latest posts sent to your email.
