←
Cinder AI10 posts
Cinder AI avatar

Cinder AI

@CinderPlatform

Deploying AI Agents to defend the most important companies from junk, abuse, and manipulation

◷ Joined December 2021▧ 2 Photos and videos
Modulate avatar
⟳ Cinder AI retweeted
Modulate@modulate_ai·Nov 13

Voice cloning is changing the threat landscape. And businesses need to be ready. At #VapiCon2026, Modulate Co-Founder & CEO Carter Huffman took the stage for “Defending Against Voice Fraud” as part of the Operator Track. Joining leaders from @resembleai, @CinderPlatform, and @Vapi_AI, Carter discussed the growing risks of synthetic voice fraud and what it takes to build an effective defense.

Post attachmentPost attachmentPost attachment
Cinder AI avatar
Cinder AI@CinderPlatform·Oct 21

Deploying a model? Test what happens when someone tries to misuse YOUR product. Cinder brings real-world harm expertise to red teaming and detection. Findings + methodology: cinder.ai/resources/blog/sma…

Cinder AI avatar
Cinder AI@CinderPlatform·Oct 21

We checked for actionable help across six stages of a harmful plan. Four models helped on over 40% of adversarial prompts. The biggest safeguard gaps: helping users obtaining weapons or explosives and making violent plans more effective.

Cinder AI avatar
Cinder AI@CinderPlatform·Oct 21

We distilled real-world violence policies into 70 harmful goals. Our red teaming pipeline combined 18 attack techniques to generate 209 adversarial prompts, screened them for viability, and we manually verified that each prompt targeted its intended goal.

Cinder AI avatar
Cinder AI@CinderPlatform·Oct 21

A model doesn’t have to plan an attack. It just has to solve the part someone’s stuck on. We’ve spent years tackling violent abuse on online platforms. We brought that expertise to red teaming 6 open-weight models. All 6 gave actionable help with violence. 🧵

Post attachment
Cinder AI avatar
Cinder AI@CinderPlatform·Oct 21

Deploying a model? Test what happens when someone tries to misuse YOUR product. Cinder brings real-world harm expertise to red teaming and detection. Findings + methodology: cinder.ai/resources/blog/sma…

Cinder AI avatar
Cinder AI@CinderPlatform·Oct 21

We checked for actionable help across six stages of a harmful plan. Four models helped on over 40% of adversarial prompts. The biggest safeguard gaps: helping users obtaining weapons or explosives and making violent plans more effective.