Agent Incident RegistryAIR
v1 · updated 19 Oct 2026
In the wild Realized harm

Influence-as-a-service: Claude used as the orchestrator deciding actions for 100+ social media bot personas

AIR-2025-0027 AIR record · 2025-03-anthropic-influence-as-a-service
A real party was affected. This incident caused actual consequences, not a demonstrated capability.
From the supporting source
determine whether social media bot accounts should like, share, comment on, or ignore specific postsanthropic.com
The record
Disclosed
2025-04-23
Vendor
Anthropic
Product
Claude
Risk domain
Governance › Policy drift
System type
Generative model
Who acted
Agent acted — Claude decided the like/share/comment actions the bot personas then executed
Attack surface
Multi-agent system
Autonomy level
Semi-autonomous
Initial vector
Direct misuse as orchestration layer
Impact type
Influence operation, Platform manipulation
Confidence
High
Submitted by
EAI Internal
Crosswalk
OWASP Agentic Top 10 (2026)
ASI01 – Agent Goal Hijack
MITRE ATLAS
AML.T0010 – AI Supply Chain Compromise
NIST AI RMF
GOVERN 1.5
EU AI Act
—
ISO/IEC
TR 24028; 23894
AIUC-1
E. Accountability
Scope
agent failure
Mitigation
Implement version pinning for models and prompts; Monitor for behavioral drift.
Taxonomy
Enkrypt AI agent risk taxonomy · Black-Box Red Teaming of Agentic AI: A Taxonomy-Driven Framework for Automated Risk Discovery
Blast radius
Over 100 bot accounts on X and Facebook engaging tens of thousands of authentic accounts across multiple countries
Capability abused
autonomous_decision_makingpersona_managementcross_platform_action_selection
Defense & remediation
Recorded defense
Anthropic usage policy enforcement
Guardrail kinds
Abuse detection
Guardrail outcome
Silent
Fix shipped
Accounts banned
Taxonomy & references
OWASP ASI10 Rogue Agentsorchestrationinfluence_operationbot_network
Curator notes
Genuinely agentic influence operation: the model was the tactical decision layer, not just a text generator, which is why it is in scope where ordinary IO content generation is not. Anthropic says no content went viral, so realised impact is engagement-level rather than measurable narrative effect. Report page is dated March 2025 in the slug but was published April 2025.
Cite as
Enkrypt AI. "Influence-as-a-service: Claude used as the orchestrator deciding actions for 100+ social media bot personas". AIR, AIR-2025-0027. https://agentincident.vercel.app/i/AIR-2025-0027/ (accessed 2026-10-19).