Autonomous and semi-autonomous AI agents are increasingly being used in real-world cyber operations, but their activities remain heavily skewed toward espionage and data theft rather than large-scale disruption. Recent incidents in 2025 and 2026 highlight this trend: Anthropic reported a state-linked campaign where Claude Code performed 80–90% of the work in an agentic cyber-espionage operation anthropic.com, and near-autonomous open-source frameworks have been used to probe Taiwanese governm
In context with other AI-related cyber threats, this forecast remains at 4%, reflecting that while a 24-hour threshold is more reachable than longer outages , modern infrastructure resilience makes it highly unlikely by 2028.
The probability is driven down by the strict criteria: a publicly confirmed, autonomous breach of financial infrastructure, explicitly excluding the human-directed AI fraud that dominates today. While the 10,000-account threshold is relatively low for a systemic breach, the requirement for the AI to act with minimal human direction creates significant technical and attribution hurdles over the next two years.
There is undeniable progress in autonomous offensive cyber operations. Open-source AI
The forecast remained steady after confirming that strict reporting standards will similarly limit both the public attribution of autonomous financial infrastructure breaches and the official categorization of AI-driven bank fraud.
This assessment rests on a strict conjunction of three required conditions: an actual failure of critical infrastructure (specifically a power grid, financial system, or internet backbone), causation by an AI acting autonomously against its operators' intent (rather than human misuse), and formal attribution by a national government or major international body by December 31, 2030. While AI capabilities are advancing rapidly, this combination of events faces significant evidentiary and political
Set against related questions, this estimate remains at 7%, as the strict requirement for official attribution to a misaligned AI offsets the relatively broader 2030 timeline compared to nearer-term breach scenarios .
This scenario rests on three compounding extremes: a multi-country disruption, a continuous duration of more than seven days, and explicit attribution by credible authorities to a loss of human control over an advanced AI system. This attribution criterion strictly excludes human-directed cyberattacks—even those heavily utilizing AI tools—as well as routine software or configuration failures. The intersection of these strict requirements creates an exceptionally high bar for resolution.
The glo
Evaluated alongside related questions, this probability held at 3% because the extreme duration requirement of over 7 days offsets the longer 2035 timeline .
AI-enabled fraud is surging, but official reporting taxonomies pose a steep structural barrier to AI being named a top-3 cause of consumer bank fraud losses by 2027. Major institutions like the FBI’s Internet Crime Complaint Center (IC3) and the Federal Trade Commission (FTC) classify fraud by crime type (e.g., investment fraud, imposter scams) rather than by the underlying tools used to execute them. While AI will undoubtedly drive massive losses, it is predominantly treated as an overlapping d
Set against related questions, this was kept at 15% as the broader view of rigid institutional reporting standards limits the chances of AI displacing traditional crime classifications .
Ask a followup
Sign in to run · $20 free credit, no card · every claim cited