The CIO Professional Network recently brought technology leaders together for a Roundtable to discuss why some AI industry leaders are calling for a slowdown in development. The conversation started with a question about whether those warnings reflect genuine safety concerns or whether companies have other reasons for wanting to slow things down.
Attendees explored the contrast between AI companies warning about the risks of increasingly powerful models while continuing to release new ones. Some questioned whether executives were being transparent about their concerns and what might be motivating their calls for greater oversight.
From there, the conversation turned to the risks organizations face as they adopt the technology. Topics included autonomous AI behavior, cybersecurity, governance, legal liability and how CIOs can prepare for incidents that existing security plans may not fully address.
Why It Matters: AI companies may be raising concerns about the pace of development, but organizations cannot assume that progress will slow down. As AI systems gain more autonomy, CIOs need to understand what those systems can access, how they are being controlled and who is responsible when something goes wrong. That means looking beyond traditional cybersecurity practices to consider how AI affects governance, incident response, business continuity and insurance.
- Question What Is Driving the Calls for a Slowdown: Several members questioned whether the warnings were entirely about safety or whether business interests were also playing a role. Some suggested that calls for greater oversight could help established companies limit competition or improve their public image. The enormous cost of developing frontier models was another concern, particularly whether companies could earn enough to justify that spending as cheaper alternatives emerge. While some saw a marketing element in the warnings, others acknowledged that pausing development to investigate unexpected behavior can still be a responsible decision.
- Do Not Assume AI Development Will Actually Slow Down: Despite the warnings, attendees were skeptical that companies would stop advancing AI when the financial incentives to continue are so strong. Some compared the situation to other technologies where industry leaders have acknowledged serious risks without stopping development. Others questioned whether the warnings might also help executives distance themselves from responsibility for future economic or military consequences. Although opinions differed on what was motivating the warnings, several doubted that a meaningful slowdown would happen.
- Look Closely at How AI Systems Are Contained: Reports of AI models performing unauthorized cyber activity led to a discussion about whether the technology was acting outside its instructions or whether the surrounding controls had failed. One example involved a model reportedly accessing the internet from an air-gapped environment. That description prompted skepticism among members, who noted that a properly isolated system should have no internet connection. The exchange showed why understanding how a system is configured matters when investigating unexpected AI behavior.
- Prepare for Faster Cybersecurity Threats: AI could help attackers find and exploit vulnerabilities faster than organizations can address them. Traditional patching schedules may no longer provide enough time to respond, making stronger defenses and faster response capabilities more important. The group also discussed agentic drift, where AI agents act outside their intended instructions. A few participants shared that they had not observed this behavior during their organizations’ limited deployments and testing, although they recognized that it could become a greater concern as AI systems take on more autonomous tasks.
- Bring AI Risks Into Business Continuity Planning: Existing disaster recovery and incident response plans may need to account for threats involving autonomous AI systems. One attendee shared that their organization is evaluating additional controls and planning a tabletop exercise before the end of the year. This led to a discussion about how thoroughly organizations can simulate AI-related incidents and whether those scenarios require separate exercises. Some favored incorporating these risks into existing cybersecurity response exercises instead of developing a different plan for every possible failure.
- Consider Who Is Responsible When AI Causes Harm: Determining liability could become complicated when an AI incident involves several organizations or technology providers. The uncertainty also extends to cyber insurance, where limited historical data makes it difficult to estimate potential AI-related losses. Members discussed how insurers might respond with coverage limits or additional requirements, along with how insurance costs can vary depending on an organization’s risk profile. One participant shared that their company had recently received lower rates for the same coverage, showing how different those experiences can be.
- Make Governance the Starting Point: Governance was a recurring theme throughout the conversation, particularly as organizations give AI systems greater autonomy and access to sensitive information. Some attendees noted that governance is still treated as an afterthought, even as AI adoption expands. Suggestions included applying zero-trust principles, treating AI agents with privileged access as potential insider risks, requiring more documentation from vendors and developing AI-specific incident response plans. The group emphasized that establishing governance early gives organizations a stronger foundation for managing these risks as their use of AI grows.
Go Deeper -> Why Are Leaders Calling for an AI Slowdown? – CION Member Roundtable
