Safety · Concept
Industry-Wide Jailbreak Severity Scoring Framework
Anthropic-led proposal with major cloud partners for standardized jailbreak severity scoring.
Forecast horizon
Standardized jailbreak severity scoring could become a baseline for AI safety compliance reporting.
Connections
Connections · 13
How this node ties into the rest of the map, and the evidence behind each link.
from · publishes 8/10
Anthropic Anthropic proposed an industry-wide jailbreak severity scoring framework together with Amazon, Microsoft, Google, and Glasswing partners.
+8 growth from · informs 6/10
Model Hypnosis: Additive Subliminal Prompt Control Additive inconspicuous prompt control creates stealthy jailbreak-like failures that severity frameworks may need to score.
+6 growth from · supports 5/10
GPT-Red: OpenAI LLM Super-Hacker for Adversarial Hardening Offensive super-hacker models and shared jailbreak severity scoring are complementary industry safety-evaluation infrastructures.
+5 growth to · informs 4/10
EU AI Office An industry-wide jailbreak severity framework could inform EU AI Office GPAI model evaluation criteria.
+4 growth from · drives 4/10
Fundamental Insecurity of LLMs to Adversarial Attacks If full hardening is impossible, standardized jailbreak severity scoring and incident response become central mitigations.
+4 growth from · drives 4/10
Fundamental Flaw Leaving LLMs Attack-Vulnerable If models are architecturally un-hardenable, industry severity scoring and external controls become more central.
+4 growth from · informs 4/10
Rules vs Character Scaling Laws for AI Safety Design Formal allocation between character shaping and rule filters informs how industry scores and invests in defenses.
+4 growth to · frames 3/10
AI-Enabled Persuasion Capabilities Jailbreak severity scoring is a governance response to AI persuasion and manipulation capabilities.
+3 growth to · informs 3/10
EU AI Act Industry jailbreak severity scoring frameworks are likely to inform AI Act compliance requirements for GPAI model providers.
+3 growth to · applies to 3/10
Claude Fable 5 / Claude Mythos 5 / Claude Opus 4.8 (Anthropic) The jailbreak severity framework was proposed alongside the redeployment of Fable 5 and its cyber safeguards.
+3 growth to · constrains 3/10
Emergent Misalignment from Narrow Fine-Tuning The jailbreak severity framework is designed to constrain and measure misalignment risks from jailbreak attacks on frontier models.
+3 growth to · supports 3/10
Science of AI Evaluation Industry-wide jailbreak severity scoring standardizes evaluation of adversarial attacks across AI systems.
+3 growth to · informs 3/10
EvalSafetyGap: LLM Evaluation-Safety Measurement Framework Industry jailbreak severity scoring addresses the evaluation-safety gap identified in EvalSafetyGap framework.
+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 →