Tag

#AI security

1 post tagged #AI security.

AI ResearchNovember 9, 20264 min read

Can poisoned training data plant a backdoor in an AI agent?

A new study says yes: poisoning 0.1% of a model's pretraining data planted a backdoor that made an AI agent run an attacker's command, and it survived clean training afterward. Here is what the researchers found.

More tags

#AI companions 16 #Lumina AI 13 #AI memory 12 #Space 6 #AI companion 5 #Human-AI teaming 5 #NASA 5 #Inventi Matrix 4 #Invention 4 #Persistent memory 4 #AI apps 3 #AI basics 3 #Frontier research 3 #How-to 3 #Living Neural Map 3 #AI collaboration 2 #AI research 2 #Buyer's guide 2 #Patents 2 #Responsible AI 2 #Voyager 2 #Wellbeing 2 #AI agents 1 #AI oversight 1 #AI personalization 1 #AI security 1 #AI transparency 1 #Astronomy 1 #Automation bias 1 #Centaur chess 1 #Character AI alternative 1 #Daily assistant 1 #Dark Matter 1 #Deep Space Network 1 #Digital wellbeing 1 #Engineering drawings 1 #ESA 1 #ExoMars 1 #Exoplanets 1 #Future of work 1 #Gravitational Lensing 1 #Gravitational Waves 1 #Intellectual property 1 #Journaling 1 #Knowledge graph 1 #Learning 1 #LISA 1 #Loneliness 1 #Mars 1 #Moon 1 #Planetary Science 1 #press release 1 #Product development 1 #Productivity 1 #Prototyping 1 #Reflection 1 #Replika alternative 1 #Saturn 1 #Technology readiness 1 #Telescopes 1 #Thinking partner 1 #thought leadership 1 #Trustworthy AI 1

#AI security

Posts

Can poisoned training data plant a backdoor in an AI agent?

More tags