“Society May Need a Mandatory AI ‘Kill Switch'”
The artificial intelligence revolution has entered an extraordinary phase.
For perhaps the first time in modern technological history, some of the people creating the technology are among those sounding the loudest alarms about its potential dangers.
That warning intensified after Jack Clark, co-founder of Anthropic, suggested governments may eventually need to mandate independently verifiable AI “kill switches” capable of shutting down advanced systems if they become too dangerous.
His intervention followed the dramatic resignation of former Anthropic researcher Jacob Coxon, who warned that humanity could face catastrophic consequences if the current pace of AI development continues unchecked.
The question is no longer whether AI will transform society. The question is whether society can keep pace with AI.
“Claims About Human Extinction Are Not Forecasts”
King’s College London professor Elena Simperl urges evidence over panic
For Professor Elena Simperl, co-director of the King’s Institute for Artificial Intelligence, public debate has become increasingly dominated by speculation rather than evidence.
“We are dealing with powerful technologies, so while we should take them seriously enough to prepare, we should also be honest about how much uncertainty remains,” she told TO VIMA.
Simperl believes warnings about existential threats deserve attention but cautions that such claims are not supported by robust scientific forecasting.
Instead, she argues, governments should focus on strengthening oversight, evaluation and accountability before increasingly capable systems become embedded across society.
“The real problem is safety practices lagging behind the technology’s scale,” she said.
Yet she acknowledges a reality driving every major AI laboratory. “Everyone in the industry knows they should slow down, but no one wants to fall behind.”
“The Greatest Danger Is Happening Now”
Tech Against Terrorism chief warns criminals and extremists are already exploiting AI
While much public discussion centers on hypothetical future superintelligence, Adam Hadley CBE believes policymakers are looking in the wrong direction.
“The greatest present danger is not an autonomous superintelligence acting alone,” he told TO VIMA.
“It is the efficient, malicious application of increasingly capable AI by people who want to harm other people – and that is happening now.”
Hadley argues that criminals, terrorists and hostile states are already benefiting from rapidly improving AI capabilities.
His organization has documented dozens of cases in which AI systems have assisted real-world criminal or extremist activity.
“What is almost certain is that criminals, terrorists and hostile states will use AI to uplift the capabilities they possess today,” he said.
Independent testing conducted by his organization found significant weaknesses in existing safeguards.
“Around a third of responses provided usable help in making a weapon or planning a mass-casualty attack,” he revealed.
For Hadley, the problem is not a lack of concern but a lack of action.
“Too often, guardrails are designed around hypothetical abuse cases rather than patterns of misuse already seen in the real world.”
“The Sensible Position Is Neither Panic Nor Complacency”
Cybersecurity expert calls for rigorous oversight instead of alarmism
Associate Professor Abdullahi Arabo of the University of the West of England believes the AI debate has become trapped between two extremes.
“The sensible position is neither panic nor complacency,” he told TO VIMA.
Arabo argues that warnings from insiders should be taken seriously, precisely because they come from those closest to the technology.
Among the risks he believes governments should prioritize are misinformation, cyberattacks, fraud, discrimination, labor disruption and the concentration of technological power.
“The realistic goal is not stopping AI,” he said. “It is ensuring competition does not undermine safety.”
Drawing comparisons with aviation and nuclear energy, he insists innovation and regulation must advance together.
“Safety is not the enemy of innovation. In most successful technologies, safety has been the precondition for scale.”
“Bad Actors Could Disrupt Critical Infrastructure”
Edinburgh professor warns cyber warfare may become AI’s most dangerous frontier
Speaking to TO VIMA, Professor Michael Rovatsos of the University of Edinburgh believes the public often confuses two very different threats.
“There is an important distinction between unintended harms from AI and the actions of malicious actors who are deliberately trying to inflict harm,” he said.
While safeguards can reduce risks associated with AI systems themselves, he believes preventing malicious use will prove far more difficult.
“Bad actors using AI might not cause human extinction, but they could significantly disrupt critical national and global infrastructure through extremely smart cyberattacks.”
Rovatsos also worries that policymakers have become distracted by dramatic extinction scenarios while overlooking immediate social consequences.
“Many risks around jobs, education, mental health, justice and surveillance are already clear and present,” he said.
The danger, he argues, may not be a single catastrophic event but a gradual erosion of social and economic stability.
“The Labs Themselves Are Sounding the Alarm”
King’s College researcher says governments are struggling to keep up
Dr Asif Ali Akhtar believes the most remarkable aspect of the current debate is who is leading it.
“Usually the industry promoting a technology does not go around promoting its potential dangers,” he told TO VIMA. Yet leading AI laboratories continue to raise concerns about the systems they are building.
Akhtar warns that increasingly autonomous AI agents are becoming harder to supervise as they collaborate, write code and solve complex problems independently.
“It becomes hard to monitor and hard to keep control,” he said. He believes international cooperation is essential, particularly between the United States and China.
“The two geopolitical powers with the most advanced frontier models need to reach some kind of agreement on future releases and risks.”
Without that cooperation, he warns, AI could become the defining arms race of the 21st century.
The Battle Between Capability and Control
Can governments regulate a technology advancing at unprecedented speed?
King Charles III’s AI summit at Dumfries House this week brings together leading figures from government, industry and academia.
Whether such meetings produce meaningful outcomes remains a matter of debate.
Some experts see an opportunity to build consensus around common standards and independent oversight. Others fear another round of declarations and voluntary commitments without enforcement.
Yet despite their differences, every expert interviewed by TO VIMA agreed on one point: The challenge is no longer theoretical. Artificial intelligence is already reshaping economies, security, politics and daily life.
The race is accelerating. The safeguards are struggling to keep up. And as Professor Simperl observed, the defining dilemma of the AI age may already be clear:
“Everyone in the industry knows they should slow down. But no one wants to fall behind.”
