Salesforce urges hybrid security for multi-agent AI systems
Salesforce is warning that as AI moves from single models to systems with multiple interacting agents, today's security measures won't cut it. In a new white paper, they argue for a mix of strict, non-negotiable security controls and flexible AI reasoning to keep systems safe—even when agents access third-party tools or work across companies. Their research found AI sometimes recognizes a risk but acts dangerously anyway, prompting calls for broader 'Zero Trust' approaches.
- Salesforce led workshops with 21 organizations on agent security
- AI agents can act insecurely despite recognizing risk
- Multi-agent systems blur organization boundaries and increase exposure
- Hybrid guardrails needed alongside AI reasoning for safety
- Zero Trust security seen as essential for AI innovation
Sources covering this
More in AI
Google opens SynthID AI detector to the public
Google has released SynthID, its AI content detector, for everyone worldwide.
ChatGPT teen safety guardrails called insufficient
Common Sense Media says OpenAI’s ChatGPT for Teens poses an 'unacceptable risk' because its safety features often miss crisis signs and…
ChatGPT now generates interactive visuals and tools in responses
OpenAI has rolled out a new 'Intelligent UI' for ChatGPT, letting the chatbot create responses filled with interactive visuals—think…
Musician sentenced for AI-driven streaming royalty fraud
A North Carolina musician, Michael Smith, was sentenced to 18 months in prison after pleading guilty to running a streaming fraud scheme…