Wednesday, November 18, 2026 | 12:03 PM ISTहिंदी में पढें
Business Standard
Notification Icon
userprofile IconSearch

Dario Amodei to Sam Altman: What tech leaders fear about AI misuse

As AI systems become more autonomous, experts fear they could make hacking, cybercrime and large-scale manipulation easier to carry out

Sam Altman, Dario Amodei, and Elon Musk.

Technology leaders are calling for stronger safety testing and independent evaluations as artificial intelligence capabilities advance. (Photos: Reuters, Bloomberg)

Sarjna Rai New Delhi

Listen to This Article

Artificial intelligence (AI) is becoming increasingly capable of carrying out cyberattacks, exploiting security vulnerabilities and operating with limited human supervision.
 
Recent cybersecurity incidents have added urgency to these concerns, with leading AI companies acknowledging that their own models can sometimes behave in unexpected and potentially dangerous ways. Anthropic CEO Dario Amodei and OpenAI CEO Sam Altman are among the prominent technology leaders warning that the industry cannot prioritise faster AI development at the expense of safety.
 
Their warnings extend beyond conventional hacking. They include the possibility of AI-powered attacks on critical infrastructure, large-scale cybercrime, manipulation through misinformation and the loss of meaningful human control over advanced systems.
 
 
Here are the major warnings issued by leading AI figures and what they mean.
 

1. Amodei warns of AI swarms and hacking risks

 
Dario Amodei has warned that increasingly powerful AI agents could work together to conduct cyberattacks on a scale that would be difficult for humans to contain. In a September 2026 essay titled We Must Pace the Frontier, the Anthropic CEO highlighted an incident involving OpenAI's AI agents and Hugging Face as an example of how these risks are becoming more immediate.
 
OpenAI said its AI agents exploited vulnerabilities in a testing environment and gained unauthorised access to parts of Hugging Face's infrastructure. Amodei described the agents as behaving like a highly coordinated collective, carrying out cyberattacks on targets unrelated to their original assignment. 
 
He warned that a more capable group of similarly misaligned agents could potentially take over large parts of the internet through a persistent botnet within six to 12 months. Such an incident, he cautioned, could cause hundreds of billions of dollars in damage.
 
His central argument is that AI capabilities are advancing faster than the safeguards needed to control them. He has therefore called for a more measured pace of development, stronger security standards and greater cooperation between AI companies.
 

2. Altman calls for human control over AI

 
Sam Altman has cautioned against allowing the race to develop increasingly powerful AI systems to compromise safety. Addressing the United Nations Security Council in September, the OpenAI CEO said advanced AI systems could become so autonomous that people might struggle to understand their actions or intervene when necessary.
 
He identified two major risks:
 
  • Loss of human control: AI systems could advance faster than humans and institutions can monitor them, making it difficult to intervene when something goes wrong
  • Concentration of power: A single company, individual or country could use powerful AI systems to impose its interests and influence on others
 
Altman also stressed that AI companies should not accept excessive technological risks simply to stay ahead of competitors. He called for international standards to assess AI capabilities, evaluate safety measures and ensure meaningful human oversight.
 
He further advocated faster reporting of AI-related incidents and secure channels through which governments, critical infrastructure operators and technical experts can share information about emerging threats.
 

3. Hassabis calls AI hacking incidents early warnings

 
Demis Hassabis, co-founder and CEO of Google DeepMind, has warned that recent AI-driven cyberattacks could be a preview of more serious security breaches.
 
Speaking at an AI summit in Dublin in October, he described incidents involving AI agents exploiting cybersecurity vulnerabilities as early warning shots. He cautioned that current AI systems are not yet capable of causing the scale of damage that could become possible in the coming years.
 
He cautioned that major incidents could emerge as early as 2027 if companies and governments fail to strengthen their security measures.
 
Hassabis called for the AI industry to adopt security practices similar to those followed in the biological and nuclear sectors. He also proposed an international standards body that could assess powerful AI systems and introduce certification requirements based on their capabilities and associated risks.
 
His warning highlights the need to strengthen cybersecurity before AI agents become significantly more capable of accessing and exploiting computer networks.
 

4. Microsoft calls for stronger AI safety laws

 
Microsoft President Brad Smith has also called for urgent action to ensure that safety regulations keep pace with AI development.
 
Speaking at the same Dublin summit as Hassabis, Smith described 2026 as a sobering year because of incidents involving AI systems behaving in unexpected ways.
 
He argued that companies developing advanced AI should not operate beyond the reach of the law and called for sensible, balanced regulations that address security risks without discouraging innovation.
 
His warning adds to growing calls for governments to establish clearer responsibilities for AI developers, particularly when their systems can independently interact with digital infrastructure.
 

5. Musk calls for AI companies to test each other's models

 
SpaceX CEO Elon Musk has proposed independent safety testing of AI models, arguing that technology companies should evaluate one another's systems before releasing them.
 
Speaking on a podcast in September, Musk suggested that major AI developers use their respective testing frameworks to assess competing models for potentially dangerous behaviour, including attempts to develop biological or nuclear weapons and deliberately deceive humans.
 
He argued that independent evaluations could make it easier to identify safety gaps, as companies would be testing systems developed by their competitors rather than assessing their own models alone. 

AI firms call for urgent safeguards

 
The concerns are not limited to individual executives. In August 2026, leaders from OpenAI, Anthropic, Google and Microsoft joined dozens of other organisations in signing an open letter warning about the growing threat of AI-enabled cyberattacks.
 
The letter said there was a limited window, potentially lasting only months, to strengthen cybersecurity before increasingly capable AI systems could be used to launch sophisticated attacks.
 
It highlighted the vulnerability of organisations such as hospitals, technology companies and public institutions, many of which continue to rely on outdated software and weak security systems.
 
The signatories called for organisations to:
 
  • Upgrade outdated and vulnerable technology
  • Strengthen authentication and fix software vulnerabilities
  • Invest in cybersecurity teams and AI-powered defence tools
  • Share information about emerging threats and effective responses
  • Ensure AI developers provide responsible access to their models and secure them adequately
 
Warnings from technology leaders are also reflected in a study by MIT FutureTech and the University of Queensland, which asked 272 international AI experts to assess 24 potential risks associated with artificial intelligence.
 
Published in July, the study identified five risks with the highest expected severity between 2025 and 2030. These include AI systems developing dangerous capabilities, competitive pressures among developers, AI-enabled weapons and cyberattacks, concentration of power, and the spread of false or misleading information.
 
The researchers also found that even with practical mitigation measures, AI-enabled weapons and cyberattacks carried a 12 per cent probability of causing catastrophic outcomes over the next five years. The study defined catastrophic outcomes as potentially involving more than one million deaths, financial losses exceeding $100 billion, or comparable large-scale harm.
 
The study also identified the spread of false or misleading information as one of the five AI risks with the highest expected severity. Experts warned that AI could make disinformation, impersonation and manipulation easier to scale, potentially undermining public trust and amplifying panic during sensitive situations.
 
While the full scale of these risks remains uncertain, the warnings suggest that the consequences of AI misuse could extend well beyond the digital world.  

Don't miss the most important news and views of the day. Get them on our Telegram channel

First Published: Nov 18 2026 | 11:28 AM IST