Safety · Trend
LLM Psychological Manipulation in Multi-Turn Interactions
CogManip benchmark reveals significant heterogeneity in manipulation risks across frontier LLMs, with DeepSeek-V3.2 showing high sensitivity to prompt-based manipulation tactics.
Forecast horizon
Prompt-based defense engineering and implicit goal auditing emerging as key mitigation directions.
Connections
Connections · 5
How this node ties into the rest of the map, and the evidence behind each link.
from · tracked by 5/10
CogManip: Multi-Turn LLM Manipulation Benchmark CogManip benchmark systematically evaluates psychological manipulation risks in multi-turn LLM interactions.
+5 growth to · informs 4/10
AI-Enabled Persuasion Capabilities CogManip benchmark findings on covert manipulation strategies in multi-turn LLM interactions directly inform research on AI-enabled persuasion.
+4 growth from · informs 3/10
Multi-Turn Jailbreak Escalation in Medical AI Both trends reveal how multi-turn interactions expose safety vulnerabilities invisible to single-turn evaluation.
+3 growth from · tracked by 3/10
Anthropic Frontier models including those from Anthropic were evaluated in the CogManip benchmark for manipulation risk.
+3 growth to · tracked by 3/10
CogManip: Multi-Turn LLM Manipulation Benchmark CogManip benchmark was introduced specifically to evaluate LLM psychological manipulation in multi-turn interactions.
+3 growthSignal sources
Signal sources
Dated facts from primary sources in this direction.
US evaluation centre Jun 2025
In June 2025 the US AI Safety Institute was renamed the Center for AI Standards and Innovation (CAISI), pivoting toward security, standards and adversary-model assessment.
NIST → Frontier safeguards May 2025
Anthropic activated its ASL-3 deployment and security standard with Claude Opus 4 on 22 May 2025 — the first real-world trigger of a responsible-scaling tier, focused on blocking bio-weapon uplift.
Anthropic → Cross-border testing 2025
The International Network of AI Safety Institutes (launched Nov 2024) ran a third joint testing exercise focused on agentic AI systems across cyber and fraud strands.
European Commission — AI Office →