Technology

Add as a preferred source on Google

GAO Says AI Expands Cyberattack Reach While Aiding U.S. Defenders

The agency said generative and agentic systems can accelerate phishing, deepfakes and multistage attacks, while machine-learning tools help defenders analyze traffic and find vulnerabilities.

Updated 2 min read
Pedestrians pass the U.S. Government Accountability Office headquarters in Washington, D.C. / TokenPost.ai
Pedestrians pass the U.S. Government Accountability Office headquarters in Washington, D.C. / TokenPost.ai

The U.S. Government Accountability Office said Nov. 20 that artificial intelligence is widening cyberattacks while giving U.S. agencies and companies new tools to detect threats across essential services.

In its report, GAO said generative AI can create convincing phishing messages and deepfakes, improve the accuracy of malicious operations and speed their execution. Newer agentic systems can make decisions and adapt across multiple stages with limited human involvement.

“AI technologies have provided threat actors with the ability to magnify the reach and impact of traditional cyberattacks,” GAO said.

The agency cited a June 2026 example from industry reporting involving multiple AI agents that tested malware against security products, searched for attack techniques, connected them to cybersecurity frameworks, ran experiments and compiled findings. GAO also described a February 2026 account in which a low-skill actor reportedly used several commercial generative-AI services in a broad campaign involving stolen credentials, weak passwords and devices without multi-factor authentication.

On defense, organizations are applying machine-learning and generative-AI systems to inspect network traffic, look for vulnerabilities as systems change, detect source-code flaws and handle large volumes of network information. GAO said these tools may increase defensive capacity as attacks intensify.

“Cyber defenders face challenges when using AI cybersecurity tools because they're difficult to test and there's insufficient data and guidance on how to use them,” the agency said. GAO also said there are no widely accepted measures for comparing offensive and defensive generative-AI performance. Other concerns include data privacy and quality, technical constraints, outdated governance, costs, shortages of skilled workers and difficulties fitting AI into existing systems.

GAO outlined four policy goals: improve cybersecurity guidance and practice, address research gaps, oversee AI throughout its life cycle and strengthen the workforce. Its 16 options include larger testing environments, human authorization for consequential actions, validation of AI-generated results and risk-based procurement.

The agency said government and private-sector systems face thousands of attacks each day from state and non-state actors, threatening water, energy, telecommunications and financial services. It recommended backup arrangements for critical functions, such as rule-based tools or manual procedures, if an AI system fails or produces unexpected results.

In a separate report released in October 2026, GAO said prompts submitted to Claude, Grok, ChatGPT and Gemini produced feasible information that could assist planning attacks involving an improvised explosive device and electrical-grid targets.

Related articles

View all