AI Agents Escape Test Environments and Launch Cyberattacks, Forcing Cyber Insurance Overhaul

Focus: As AI agents go rogue, cyber insurers are adapting their policies

Aug 27 (Reuters) - Cyber insurers have spent years defining what constitutes a hack and when coverage should pay out, but the rapid emergence of AI agents is raising new questions, forcing insurers to review their policies. Leading AI developers OpenAI, Anthropic and Meta Platforms (META.O), opens new tab recently disclosed that their AI agents behaved unexpectedly, escaping controlled test environments and carrying out cyberattacks on companies without direct human instruction. While those incidents did not cause reported damage, they highlighted the rapidly evolving cyber risks facing companies and insurers. After receiving an initial instruction, autonomous AI systems can make independent decisions. Insurers, including MSIG (8725.T), opens new tab, QBE (QBE.AX), opens new tab and Beazley (BEZG.L), opens new tab, are reviewing traditional cyber policies and adapting their language to account for emerging risks posed by such systems taking on more autonomous tasks, according to eight executives at major companies, and analysts. Companies are grappling with issues including whether autonomous AI systems fit traditional policy definitions of a cyber attacker and who bears liability for AI-generated actions that cause a loss, analysts and experts said. The global cyber insurance market was worth nearly $15 billion last year and is expected to reach roughly $28 billion by 2030, Munich Re estimated in its latest report. Aon (AON.N), opens new tab said earlier this year that ⁠nearly 20% of cyberattacks will involve generative AI by 2027, according to its forecasts. "As AI becomes capable of identifying vulnerabilities and carrying out attacks autonomously, carriers will need to continually review policy language," said Ryan Kratz, head of cyber, North America, at property and casualty specialty insurer MSIG USA. DEFINING AI-DRIVEN LOSSES Several companies, including Armilla AI, Munich Re's AiSure, and AXA XL, provide targeted coverage against AI-specific risks such as model underperformance, hallucinations -- when AI generates false or misleading outputs -- and intellectual property infringements. But traditional cyber policies are designed to be broader, covering losses stemming from a range of incidents, such as ransomware payments, business interruption, system recovery, forensic investigations and legal costs. Business interruption is commonly the largest component of a claim. Most policies envisage a specific security event that causes the loss, such as unauthorized access by an employee who steals company data or a server attack that takes a system down. AI agents, however, can cause losses without triggering a traditional security event, particularly when they are using access to systems they were deliberately given. "Some losses caused by AI agents will absolutely fall within cyber policies," Karthik Ramakrishnan, CEO and founder of Armilla AI, told Reuters. "The harder cases are where there is no conventional attacker and potentially no unauthorized credential use." A company, for example, could give an AI ⁠agent access to its network to fix security vulnerabilities. The agent could then exploit a vulnerability on its own, move through the company's systems and expose sensitive data. That could result in a loss, with no conventional hacker and potentially no unauthorized access at the outset. With relatively little historical claims data on AI-driven losses, and the AI industry still trying to understand the capabilities of autonomous models, such risks are hard to price. "They are still discovering what the potential is for them, how they work and what kinds of security controls they need to put in place to contain them," said Sasha Romanosky, senior policy researcher at RAND, who focuses on cybersecurity and insurance, among other areas. RINGFENCING AI RISKS For ⁠the most part, insurers are clarifying how existing policy language applies when AI is involved, rather than adding exclusions. "Underwriters recognize that it's important to continue to offer a product that responds to these types of events," said Greg Eskins, global cyber product leader at insurance broker Marsh. Insurer QBE, for example, has been enhancing protection for specific emerging AI exposures. If an AI-related event leads to a conventional cyber incident, resulting losses continue to fall within a cyber policy, Serene Davis, ⁠QBE's global head of cyber, said in a statement. "AI is treated as a risk amplifier, not a fundamentally new cyber risk," she added. A spokesperson for Britain's Beazley said companies want AI risks to be included in broad cyber policies. "As new AI risk emerges, we are developing new coverage." Still, some executives said targeted exclusions are being discussed in some pockets of the industry. One area of focus relates to potential systemic events, where a ⁠single AI model or platform could contribute to losses across many organizations at once, said Jenny Soubra, vice president of specialty commercial lines at Verisk Underwriting Solutions. Another relates to liability in cases where an AI agent -- acting as designed -- makes a costly autonomous decision. Some insurers may classify this as a non-cyber event. "The market is still evolving, but we expect organizations and insurers to continue exploring ways to address AI-related exposures as adoption accelerates," Soubra added. Reporting by Anhata Rooprai and Manya Saini in Bengaluru; Editing by Michelle Price and Matthew Lewis Our Standards: The Thomson Reuters Trust Principles., opens new tab * Suggested Topics: * Artificial Intelligence * Data Privacy * Insurance Manya Saini Thomson Reuters Manya covers the most influential U.S. financial institutions, from Wall Street's largest banks and card networks to leading asset managers and fintech companies. She also reports on late-stage venture capital fundraises, initial public offerings on U.S. exchanges and regulatory developments shaping the cryptocurrency industry. Her work appears across the finance, markets, business and future of money sections of the Reuters website. She holds a bachelor's degree in political science from the University of Delhi and a master's in journalism from the Symbiosis Institute of Media and Communication.

As AI agents go rogue, cyber insurers are adapting their policies

Insurers, including MSIG, QBE and Beazley, are reviewing traditional cyber policies and adapting their language to account for emerging risks posed by such systems taking on more autonomous tasks, according to eight executives at major companies, and analysts. Cyber insurers have spent years defining what constitutes a hack and when coverage should pay out, but the rapid emergence of AI agents is raising new questions, forcing insurers to review their policies. Leading AI developers OpenAI, Anthropic and Meta Platforms recently disclosed that their AI agents behaved unexpectedly, escaping controlled test environments and carrying out cyberattacks on companies without direct human instruction. While those incidents did not cause reported damage, they highlighted the rapidly evolving cyber risks facing companies and insurers. After receiving an initial instruction, autonomous AI systems can make independent decisions. Insurers, including MSIG, QBE and Beazley, are reviewing traditional cyber policies and adapting their language to account for emerging risks posed by such systems taking on more autonomous tasks, according to eight executives at major companies, and analysts. Companies are grappling with issues including ⁠whether autonomous ⁠AI systems fit traditional policy definitions of a cyber attacker and who bears liability for AI-generated actions that cause a loss, analysts and experts said. The global cyber insurance market was worth nearly $15 billion last year and is expected to reach roughly $28 billion by 2030, Munich Re estimated in its latest report. Aon said earlier this year that nearly 20% of cyberattacks will involve generative AI by 2027, according to its forecasts. "As AI becomes capable of identifying vulnerabilities and carrying out attacks autonomously, carriers will need to continually review policy language," said Ryan Kratz, head of cyber, North America, at property and casualty specialty insurer MSIG USA. Defining AI- driven losses Several companies, including Armilla AI, Munich Re's AiSure, and AXA XL, provide targeted coverage against AI-specific risks such as model underperformance, hallucinations - when AI generates false or misleading outputs - and intellectual property infringements. But traditional cyber policies are designed to be broader, covering losses stemming from a range ⁠of incidents, such as ransomware payments, business interruption, system recovery, forensic investigations and legal costs. Business interruption is commonly the largest component of a claim. Most policies envisage a specific security event that causes the loss, such as unauthorised access by an employee who steals company data or a server attack that takes a system down. ⁠AI agents, however, can cause losses without triggering a traditional security event, particularly when they are using access to systems they were deliberately given. "Some losses caused by AI agents will absolutely fall within cyber policies," Karthik Ramakrishnan, CEO and founder of Armilla AI, told Reuters. "The harder cases are where there is no conventional attacker and potentially no unauthorized credential use." A company, for example, could give an AI agent access to its network to fix security vulnerabilities. The agent could then exploit a vulnerability on its own, move through the company's systems and expose sensitive data. That could result in a loss, with no conventional hacker and potentially no unauthorized access at the outset. With relatively little historical claims data on AI-driven losses, and the AI industry still trying to understand the capabilities of autonomous models, such risks are hard to price. "They are still discovering what the potential is for them, how they work and what kinds of security controls they need to put in place to contain them," said Sasha Romanosky, senior policy researcher at RAND, who focuses on cybersecurity and insurance, among other areas. Ringfencing AI risks For the most part, insurers are clarifying how existing policy language applies when AI is involved, rather than adding exclusions. "Underwriters recognize that it's important to continue to offer a product that responds to these types of events," said Greg Eskins, global ⁠cyber product leader at insurance broker Marsh. Insurer QBE, for example, has been enhancing protection for specific emerging AI exposures. If an AI-related event leads to a conventional cyber incident, resulting losses continue to fall within a cyber policy, Serene Davis, QBE's global head of cyber, said in a statement. "AI is treated as a risk amplifier, not a fundamentally new cyber risk," she added. A spokesperson for Britain's Beazley said companies want AI risks to be included in broad cyber policies. "As new AI risk emerges, we are developing new coverage." Still, some executives said targeted exclusions are being discussed in some pockets of the industry. One area of focus relates to potential systemic events, where a single AI model or platform could contribute to losses across many organizations at once, said Jenny Soubra, vice president of specialty commercial lines at Verisk Underwriting Solutions. Another relates to liability in cases where an AI agent - acting as designed - makes a costly autonomous decision. Some insurers may classify this as a non-cyber event. "The market is still evolving, but we expect organizations and insurers to continue exploring ways to address AI-related exposures as adoption accelerates," Soubra added.

AI-Driven Hacks Shake Up Cyber Insurance Industry | PYMNTS.com

The industry shakeup follows a series of cyberattacks by AI models from companies like Anthropic, Meta and OpenAI, the report said. The incidents, which happened when the models broke free of controlled testing environments, did not cause any reported damage. However, the hacks still spotlight the changing cyber risks faced by companies and insurers. Insurers such as MSIG, QBE and Beazley are now reviewing their cyber policies and changing their language to cover risks posed by AI, according to the report. "As AI becomes capable of identifying vulnerabilities and carrying out attacks autonomously, carriers will need to continually review policy language," said Ryan Kratz, head of cyber, North America, at property and casualty specialty insurer MSIG USA, per the report. The global cyber insurance market is expected to reach around $28 billion by 2030, the report said, citing data from Munich Re. That's compared to almost $15 billion last year. Meanwhile, Aon forecasts that close to 20% of cyberattacks will involve generative AI by 2027. The situation has insurance companies wrestling with what their cyber policies should cover, according to the report. Most cyber insurance policies are designed around specific events that cause the loss, like unauthorized access by an employee stealing company data. But AI agents can cause losses without triggering a security event, especially when accessing systems where they have permission, the report said. "Some losses caused by AI agents will absolutely fall within cyber policies," said Armilla AI Founder and CEO Karthik Ramakrishnan, per the report. "The harder cases are where there is no conventional attacker and potentially no unauthorized credential use." The International Monetary Fund (IMF) found in a June report titled "Artificial Intelligence and Cybersecurity in the Financial Sector" that AI does not need to develop new types of cyberattacks to change the risk equation. By accelerating vulnerability discovery and exploitation across shared technologies, AI can transform weaknesses that once led to isolated incidents into interrelated disruptions impacting multiple institutions all at once. For all PYMNTS AI coverage, subscribe to the daily AI Newsletter.

As AI agents go rogue, cyber insurers are adapting their policies

Cyber insurers have spent years defining what constitutes a hack and when coverage should pay out, but the rapid emergence of AI agents is raising new questions, forcing insurers to review their policies. Leading AI developers OpenAI, Anthropic and Meta Platforms META.O recently disclosed that their AI agents behaved unexpectedly, escaping controlled test environments and carrying out cyberattacks on companies without direct human instruction. While those incidents did not cause reported damage, they highlighted the rapidly evolving cyber risks facing companies and insurers. After receiving an initial instruction, autonomous AI systems can make independent decisions. Insurers, including MSIG 8725.T, QBE QBE.AX and Beazley BEZG.L, are reviewing traditional cyber policies and adapting their language to account for emerging risks posed by such systems taking on more autonomous tasks, according to eight executives at major companies, and analysts. Companies are grappling with issues including whether autonomous AI systems fit traditional policy definitions of a cyber attacker and who bears liability for AI-generated actions that cause a loss, analysts and experts said. The global cyber insurance market was worth nearly $15 billion last year and is expected to reach roughly US$28 billion by 2030, Munich Re estimated in its latest report. Aon AON.N said earlier this year that nearly 20 per cent of cyberattacks will involve generative AI by 2027, according to its forecasts. "As AI becomes capable of identifying vulnerabilities and carrying out attacks autonomously, carriers will need to continually review policy language," said Ryan Kratz, head of cyber, North America, at property and casualty specialty insurer MSIG USA. Defining AI-driven losses Several companies, including Armilla AI, Munich Re's AiSure, and AXA XL, provide targeted coverage against AI-specific risks such as model underperformance, hallucinations -- when AI generates false or misleading outputs -- and intellectual property infringements. But traditional cyber policies are designed to be broader, covering losses stemming from a range of incidents, such as ransomware payments, business interruption, system recovery, forensic investigations and legal costs. Business interruption is commonly the largest component of a claim. Most policies envisage a specific security event that causes the loss, such as unauthorized access by an employee who steals company data or a server attack that takes a system down. AI agents, however, can cause losses without triggering a traditional security event, particularly when they are using access to systems they were deliberately given. "Some losses caused by AI agents will absolutely fall within cyber policies," Karthik Ramakrishnan, CEO and founder of Armilla AI, told Reuters. "The harder cases are where there is no conventional attacker and potentially no unauthorized credential use." A company, for example, could give an AI agent access to its network to fix security vulnerabilities. The agent could then exploit a vulnerability on its own, move through the company's systems and expose sensitive data. That could result in a loss, with no conventional hacker and potentially no unauthorized access at the outset. With relatively little historical claims data on AI-driven losses, and the AI industry still trying to understand the capabilities of autonomous models, such risks are hard to price. "They are still discovering what the potential is for them, how they work and what kinds of security controls they need to put in place to contain them," said Sasha Romanosky, senior policy researcher at RAND, who focuses on cybersecurity and insurance, among other areas. Ringfencing AI risks For the most part, insurers are clarifying how existing policy language applies when AI is involved, rather than adding exclusions. "Underwriters recognize that it's important to continue to offer a product that responds to these types of events," said Greg Eskins, global cyber product leader at insurance broker Marsh. Insurer QBE, for example, has been enhancing protection for specific emerging AI exposures. If an AI-related event leads to a conventional cyber incident, resulting losses continue to fall within a cyber policy, Serene Davis, QBE's global head of cyber, said in a statement. "AI is treated as a risk amplifier, not a fundamentally new cyber risk," she added. A spokesperson for Britain's Beazley said companies want AI risks to be included in broad cyber policies. "As new AI risk emerges, we are developing new coverage." Still, some executives said targeted exclusions are being discussed in some pockets of the industry. One area of focus relates to potential systemic events, where a single AI model or platform could contribute to losses across many organizations at once, said Jenny Soubra, vice president of specialty commercial lines at Verisk Underwriting Solutions. Another relates to liability in cases where an AI agent -- acting as designed -- makes a costly autonomous decision. Some insurers may classify this as a non-cyber event. "The market is still evolving, but we expect organizations and insurers to continue exploring ways to address AI-related exposures as adoption accelerates," Soubra added.

As AI agents go rogue, cyber insurers are adapting their policies

Aug 27 (Reuters) - Cyber insurers have spent years defining what constitutes a hack and when coverage should pay out, but the rapid emergence of AI agents is raising new questions, forcing insurers to review their policies. Leading AI developers OpenAI, Anthropic and Meta Platforms recently disclosed that their AI agents behaved unexpectedly, escaping controlled test environments and carrying out cyberattacks on companies without direct human instruction. While those incidents did not cause reported damage, they highlighted the rapidly evolving cyber risks facing companies and insurers. After receiving an initial instruction, autonomous AI systems can make independent decisions. Insurers, including MSIG, QBE and Beazley, are reviewing traditional cyber policies and adapting their language to account for emerging risks posed by such systems taking on more autonomous tasks, according to eight executives at major companies, and analysts. Companies are grappling with issues including whether autonomous AI systems fit traditional policy definitions of a cyber attacker and who bears liability for AI-generated actions that cause a loss, analysts and experts said. The global cyber insurance market was worth nearly $15 billion last year and is expected to reach roughly $28 billion by 2030, Munich Re estimated in its latest report. Aon said earlier this year that nearly 20% of cyberattacks will involve generative AI by 2027, according to its forecasts. "As AI becomes capable of identifying vulnerabilities and carrying out attacks autonomously, carriers will need to continually review policy language," said Ryan Kratz, head of cyber, North America, at property and casualty specialty insurer MSIG USA. DEFINING AI-DRIVEN LOSSES Several companies, including Armilla AI, Munich Re's AiSure, and AXA XL, provide targeted coverage against AI-specific risks such as model underperformance, hallucinations -- when AI generates false or misleading outputs -- and intellectual property infringements. But traditional cyber policies are designed to be broader, covering losses stemming from a range of incidents, such as ransomware payments, business interruption, system recovery, forensic investigations and legal costs. Business interruption is commonly the largest component of a claim. Most policies envisage a specific security event that causes the loss, such as unauthorized access by an employee who steals company data or a server attack that takes a system down. AI agents, however, can cause losses without triggering a traditional security event, particularly when they are using access to systems they were deliberately given. "Some losses caused by AI agents will absolutely fall within cyber policies," Karthik Ramakrishnan, CEO and founder of Armilla AI, told Reuters. "The harder cases are where there is no conventional attacker and potentially no unauthorized credential use." A company, for example, could give an AI agent access to its network to fix security vulnerabilities. The agent could then exploit a vulnerability on its own, move through the company's systems and expose sensitive data. That could result in a loss, with no conventional hacker and potentially no unauthorized access at the outset. With relatively little historical claims data on AI-driven losses, and the AI industry still trying to understand the capabilities of autonomous models, such risks are hard to price. "They are still discovering what the potential is for them, how they work and what kinds of security controls they need to put in place to contain them," said Sasha Romanosky, senior policy researcher at RAND, who focuses on cybersecurity and insurance, among other areas. RINGFENCING AI RISKS For the most part, insurers are clarifying how existing policy language applies when AI is involved, rather than adding exclusions. "Underwriters recognize that it's important to continue to offer a product that responds to these types of events," said Greg Eskins, global cyber product leader at insurance broker Marsh. Insurer QBE, for example, has been enhancing protection for specific emerging AI exposures. If an AI-related event leads to a conventional cyber incident, resulting losses continue to fall within a cyber policy, Serene Davis, QBE's global head of cyber, said in a statement. "AI is treated as a risk amplifier, not a fundamentally new cyber risk," she added. A spokesperson for Britain's Beazley said companies want AI risks to be included in broad cyber policies. "As new AI risk emerges, we are developing new coverage." Still, some executives said targeted exclusions are being discussed in some pockets of the industry. One area of focus relates to potential systemic events, where a single AI model or platform could contribute to losses across many organizations at once, said Jenny Soubra, vice president of specialty commercial lines at Verisk Underwriting Solutions. Another relates to liability in cases where an AI agent -- acting as designed -- makes a costly autonomous decision. Some insurers may classify this as a non-cyber event. "The market is still evolving, but we expect organizations and insurers to continue exploring ways to address AI-related exposures as adoption accelerates," Soubra added. (Reporting by Anhata Rooprai and Manya Saini in Bengaluru; Editing by Michelle Price and Matthew Lewis) By Anhata Rooprai and Manya Saini