PPhying中文
Curated security research, vulnerabilities, advisories and tools for practitioners.

AI Security Testing Is the Maturity Milestone for Modern Organizations

hackerone.com | blog | #ai-security | #prompt-injection | #jailbreak | #security-testing | #model-agency | #ai-vulnerabilities | #shai-hulud
📄 Raw .md⬇️ Download .md🔗 Original ↗

Summary

Valid AI-related security reports have tripled since 2024, led by prompt injection, insecure tool design, and excessive model agency—patterns seen in recent Claude jailbreaks and Shai-Hulud 2.0.
Published
Collected

Related coverage

blog · hackerone.com
Measuring The Fix: How We Evaluate Frontier Models on Remediation Quality
HackerOne tested frontier models on security fix quality and found the judge mattered more than the model. See how CTEM remediation earns engineer trust.
blog · hackerone.com
Shai-Hulud 2.0: What Security Leaders Need to Do Now
Shai-Hulud 2.0 is a self-replicating npm worm stealing secrets from dev and CI/CD environments; with 1,000+ packages and 27,000+ repos hit, rotate credentials and disable lifecycle scripts.
blog · hackerone.com
AI Security 101: What is AI Red Teaming?
AI red teaming stress-tests AI systems against adversarial attacks, targeting jailbreaks, prompt injection and over-permissioned agents so flaws are fixed before attackers can exploit them.
blog · hackerone.com
LLMs adversarial testing with HackerOne: Join the CTF!
HackerOne and Hack The Box launch a 10-day AI red teaming CTF built from real engagement scenarios: indirect prompt injection, safety filter bypasses, and privilege escalation via language alone.
blog · hackerone.com
Rethinking Risk in the Age of AI
Generative AI is expanding attack surfaces with prompt injection, jailbreaks, and data poisoning. Anthropic's HackerOne jailbreak challenge shows how crowdsourced testing validates AI safety.
blog · hackerone.com
How Anthropic’s Jailbreak Challenge Put AI Safety Defenses to the Test
Anthropic and HackerOne ran an AI red teaming challenge on a demo Claude 3.5 Sonnet to validate Constitutional Classifiers that block CBRN-related jailbreaks, awarding $55,000 to four teams.
Back