Recorded Future Finds AI Mostly Amplifies Familiar Cyberattacks
Summary
Recorded Future analyzed 103 AI-enabled incidents from July 2025 to July 2026. The report finds AI mainly accelerates familiar attacks, while fully autonomous operations remain rare; AI developer tools are a prominent target.
Key points
- Insikt Group counted 103 documented incidents in which AI was used as an attack vector, tool, or target; 32 involved AI in malware or offensive-operation orchestration.
- The report says compromised AI development tools and packages have enabled credential theft, lateral movement, data theft, and persistence across downstream environments.
- AI-themed lures, including fake product pages and phishing, exploit interest in AI to deliver malware or steal cryptocurrency.
- State-sponsored actors mostly use AI to assist malware development, obfuscation, and targeting; one reported campaign against Taiwanese government entities involved up to eight AI agents.
- Threat actors use AI to personalize phishing and analyze stolen data for extortion; the report says this can speed and tailor extortion demands.
- Fully autonomous attacks remain uncommon in the observed incidents; the report cites operational risks, cost, and the continued need for human oversight as constraints.
- Recommended measures include limiting agent access, requiring human approval for consequential actions, monitoring credentials, and auditing third-party AI dependencies.
Article Details
- Publisher
- Recorded Future
- Report Period
- 2025-07-31 to 2026-07-31
- Scope
- AI-related malicious campaigns documented in the Recorded Future intelligence graph, including AI as an attack vector, tool, or target.
- Sample Size
- 103 incidents
- Key Statistics
- 103 incidents were identified between 2025-07-31 and 2026-07-31.
- 32 of the 103 incidents involved AI in malware or in orchestrating offensive cyber operations; the other 71 were classified into four AI-abuse categories.
- 9 of the 32 incidents in the AI Malware Maturity Model dataset were attributed to state-sponsored intelligence collection.
- Roughly one-third of incidents across both datasets were financially motivated.
- Recommendations
- Reassess organizational risk profiles in light of AI-driven automation, faster extortion timelines, and related threats.
- Limit AI agents’ access, isolate them from sensitive systems, require human approval for consequential actions, and maintain independent monitoring and shutdown capabilities.
- Monitor infostealer logs and dark web channels for compromised commercial AI subscriptions, developer API keys, and corporate OAuth tokens.
- Audit third-party AI dependencies, code-generation plugins, and open-weight model repositories for malicious command injection or backdoors.
MITRE ATT&CK
Threat Actors
Malware
Vendors
Anthropicand exploiting vulnerabilities are a recent development: AI-assisted vulnerability discovery predates Anthropic’s Mythos model, but its release provided a prominent public demonstration of these capabilities andRecorded FutureMost of the state-sponsored activity in this report falls within Recorded Future’s AI Malware Maturity Model Level 2–3 (AI-assisted malware development), with one notable exception: a campaign against Taiwanese
Products
Claude CodeThe May expansion of the Shai-Hulud supply-chain worm similarly targeted Claude Code to establish persistence and spread.Identity IntelligenceUse Recorded Future’s Identity Intelligence to detect exposed credentials.Mythosvulnerabilities are a recent development: AI-assisted vulnerability discovery predates Anthropic’s Mythos model, but its release provided a prominent public demonstration of these capabilities and coincided withOpenClawAttackers are spoofing the landing pages of popular AI developer tools, standing up fake utility pages, and impersonating products such as OpenClaw with fraudulent wallet pages designed to steal crypto assets.
Countries
FranceOne operation targeting French pensioners used stolen identity data to automate highly targeted phishing messages, accelerating the threat actors’ capabilities to deliver convincing content at scale.PakistanAI tools across a variety of operational functions, including malware development. In addition, a Pakistan-based state-sponsored actor used LLMs to produce “vibeware” capable of “living off the land” inRussiaactors are using AI at Level 2 (Adopting) or Level 3 (Optimizing). Researchers have observed Russian threat actors incorporating AI tools across a variety of operational functions, including malwareTaiwanModel Level 2–3 (AI-assisted malware development), with one notable exception: a campaign against Taiwanese government entities that employed up to eight AI agents with limited human intervention.