WitnessAI Broadcasts Automated Pink-Teaming and Subsequent-Technology AI Firewall Safety for Enterprise LLMs and AI Functions

New AI Safety Capabilities Check and Harden LLMs In opposition to Actual-World Threats Whereas Imposing Safety and Governance Insurance policies Throughout Dwell Use

Following 1 / 4 of file gross sales, WitnessAI, the safe AI enablement chief, introduced two new merchandise that safe enterprise LLMs and AI functions by way of automated pink teaming and behavioral runtime safety. Collectively, they supply world enterprises with automated capabilities for testing, hardening, and defending enterprise AI fashions, in addition to the apps and brokers that use them.

• Witness Assault is an automatic red-teaming software designed to proactively establish vulnerabilities in giant language fashions (LLMs) earlier than they’re deployed. It makes use of simulated assault methods akin to multimodal assaults, multi-step jailbreaks, complete fuzzing, API integration, and reinforcement-learning assaults, to generate artificial prompts that stress-test the mannequin’s defenses. With Witness Assault, builders can simply establish and deal with weak spots of their fashions’ personal defenses earlier than they’re rolled out.
• Witness Shield is a next-generation AI firewall that gives behavioral runtime protection towards mannequin assaults akin to immediate injections and jailbreaks. It additionally offers LLM defenses akin to behavioral immediate filtering, intention-based response management, actual time knowledge redaction, risk detection, dangerous response, toxicity filtering, and mannequin id enforcement, all delivered by way of a easy developer API. Witness Shield consists of new protection methods honed from two years’ value of artificial immediate knowledge and steady conversational assaults. Because of this, the answer offers industry-leading detection of mannequin assaults, with over 99% efficient identification of immediate injections. Witness Shield’s defenses present standardized safety throughout fashions, and work with over 100 kinds of LLMs.

Additionally Learn: AiThority Interview with Suzanne Livingston, Vice President, IBM Watsonx Orchestrate Agent Domains

With the introductions of Witness Assault and Witness Shield, the WitnessAI platform now addresses a broad set of enterprise AI safety and compliance necessities, together with secure AI growth, compliant worker AI utilization, and safe mannequin and app runtime operations.

“Enterprises don’t wish to purchase 5 completely different merchandise to make sure their staff and clients can use AI safely,” stated Rick Caccia, WitnessAI CEO. “With the introduction of Witness Assault, enterprises can now guarantee automated testing and hardening of their internally-developed fashions, apps, and brokers. Witness Shield provides even higher defenses towards mannequin assaults, and is already in buyer evaluations to exchange previously-deployed AI firewall options from legacy safety suppliers.”

Additionally Learn: C-Gen.AI Emerges from Stealth to Finish Infrastructure Limitations Affecting AI Workloads