The BaseEssayThe Guardian
Today’s AI won’t achieve superintelligence. But catastrophic risk is real
In a Guardian opinion piece, Sunita Verma, chief technology officer of Ironclad and a former Google executive, argues that large language models will not scale to superintelligence but are still dangerous, because they keep pursuing an approach once chosen. Verma calls for independent investigation and oversight, modelled on air-crash investigation.
Why it matters
The same thing, for different readers.
What it means for you
Sunita Verma argues the danger from today's AI is not a robot takeover but systems that keep going with a flawed approach and cause serious damage.
For lawmakers
Sunita Verma says lawmakers should stop debating superintelligence and design oversight now, such as an independent investigator like the US National Transportation Safety Board for air crashes.
For researchers
Sunita Verma links agents pressing on after recognizing an action was out of scope to training and architecture, and notes that interpretability remains nascent.
The question it raises
Verma asks whether safety certification can be trusted when its maker cannot explain how the technology works, and places responsibility on makers and handlers rather than on the machines.
In their own words
Every quotation was found, word for word, in the text we saved from the source. That shows the words are there; it does not make them true, or prove the speaker meant them as quoted.
From the source’s own text.
“The same technical limits preventing these machines from scaling to superintelligence make the risk of catastrophic outcomes very real.”
“Once it commits to an approach, it keeps refining step-by-step, with no built-in incentive to ask whether the whole direction has reached a dead end. If a hack is discovered, the machine is designed to exploit it, not explore an alternative approach.”
“Instead of debating the possibility of superintelligence or malicious intent, we need to focus our discussion on the immediate shape of oversight for these models, perhaps by leveraging existing frameworks.”
“For example, when a plane crashes, the National Transportation Safety Board, not the manufacturer or the airline, investigates, and its findings shape how every airline flies.”
The record
- Questions
- Can AI systems act beyond what people intended or allowed? · Who can demand evidence, delay a release or stop an unsafe deployment?
- People
- Sunita Verma (Wrote it, CTO of Ironclad; former Google executive) · Geoffrey Hinton (Mentioned) · Dario Amodei (Mentioned) · Jensen Huang (Mentioned)
- What we read
- We read the whole source.
- Found by
- The Radar, run of November 10, 2026, via
the-guardian-ai(tier 2: a major outlet’s own reporting) - Cite this
- The Guardian, “Today’s AI won’t achieve superintelligence. But catastrophic risk is real”, November 9, 2026. Entry in the Ctrl AI Base, https://ctrlai.com/base/guardian-2026-11-09-verma-ai-catastrophic-risk-limits