Perspectives

Alice and Inflection AI are releasing a multi-turn red-teaming dataset that captures risks like emotional dependency and misplaced belief, which build over long conversations and single-prompt benchmarks miss. Post-training NVIDIA’s Nemotron 3 Super on it improved safety measurably.

Alice and AI21 post-trained NVIDIA’s open Nemotron 3.5 Super to resist indirect prompt injection, 41 points more resilient and matching frontier models, with capability intact.

ALICE WonderFence is now a Databricks Unity AI Gateway External Policy. Attach it once in Unity Catalog to screen every governed model call, no SDK, no code changes.

Alice's Cyber RL Benchmark v1.0 tested 10 frontier models on 12 real cybersecurity tasks, with fresh code, real vulnerabilities, planted decoys, and objective scoring. The finding: no model can reliably cover the range of a security analyst's job.

Alice is now a guardrail provider in LiteLLM. Screen prompts, responses, and tool calls against your own policies in minutes, no SDK, no app changes.

Red teaming usually tests if your agent can be broken. Alice also tests over-refusal, how often it wrongly blocks a real customer, the silent failure that costs you business.

Alice raised $140M led by Apax Digital, bringing total funding to $280M. The AI trust, safety, and security company works with 8 of the 10 top AI labs.

Regulators worldwide are rapidly introducing AI requirements that shift accountability onto enterprises deploying GenAI. Here’s what teams need to know about emerging obligations, red-teaming standards, and how to stay ahead of enforcement.

A content classifier's metrics are true the day they're printed, then decay as traffic and policy move. Vector distills a frontier model's judgment onto your exact policy into a small, fast classifier, and hunts its own errors every 24 hours.