Alice AI Red Teaming
Expert-led and automated red teaming of foundation models across text, image, audio and video.
Offered by AliceRed teaming and evaluation
About Alice AI Red Teaming
Its Evaluations & Red Team offering combines expert-led and automated red-teaming to stress-test models across text, image, audio, and video.
Alice states that it has worked for over eight years with top-10 tech platforms on trust and safety across extreme harms, including child safety, CBRNE, deception, political bias, prompt injections, and data exfiltration. The company describes a native speaker network across dozens of languages, with support for over 100 languages. Alice describes its turnaround as days or weeks.
Drafted from alice.io · not yet confirmed by Alice
What they do
Alice (formerly ActiveFence) offers pre-launch and production evaluation to foundation model labs: red teaming, safety benchmarking, model hardening and compliance alignment. It combines expert-led manual red teaming by subject matter experts with automated testing, and draws on what it describes as eight years of trust and safety work for large platforms and a native speaker network across dozens of languages.
Harm areas it names include child safety, CBRNE, deception and political bias, alongside security issues such as prompt injection and data exfiltration. For Amazon's Nova Premier it ran manual, blind evaluations with experts writing adversarial prompts across Amazon's eight Responsible AI categories. It says turnaround is days or weeks.
Reviews
Be the first to review it
AnonymousKnow a team weighing Alice AI Red Teaming? Send them this page.