#llm-security
#llm-security
[ follow ]
#prompt-injection #mitigation #data-exfiltration #copilot-personal #url-exfiltration #kubernetes #sandboxing
#prompt-injection
Information security
fromArs Technica
4 months agoNew attack on ChatGPT research agent pilfers secrets from Gmail inboxes
Prompt injections remain largely unpreventable, forcing LLM providers to rely on reactive, channel-blocking mitigations that require explicit user consent to prevent data exfiltration.
Information security
fromInfoQ
3 weeks agoOpen-Source Agent Sandbox Enables Secure Deployment of AI Agents on Kubernetes
Agent Sandbox provides a secure, stateful, declarative Kubernetes API to run isolated, persistent, single-pod environments for safely executing untrusted LLM-generated code and stateful workloads.
fromArs Technica
2 months agoCritics scoff after Microsoft warns AI feature can infect machines and pilfer data
Microsoft's warning on Tuesday that an experimental AI Agent integrated into Windows can infect devices and pilfer sensitive user data has set off a familiar response from security-minded critics: Why is Big Tech so intent on pushing new features before their dangerous behaviors can be fully understood and contained? As reported Tuesday, Microsoft introduced Copilot Actions, a new set of "experimental agentic features"
Information security
Startup companies
fromBusiness Insider
3 months agoThis AI startup helps developers safely and cheaply build on top of LLMs from OpenAI and Anthropic. Read its pitch deck.
Requesty centralizes developer access to multiple AI providers, tightening security, enforcing governance, and optimizing costs across large language models.
fromTheregister
4 months agoAI can't stop the sprint to adopt hot tech without security
Ollama provides a framework that makes it possible to run large language models locally, on a desktop machine or server. Cisco decided to research it because, in the words of Senior Incident Response Architect Dr. Giannis Tziakouris, Ollama has "gained popularity for its ease of use and local deployment capabilities." Talos researchers used the Shodan scanning tool to find unsecured Ollama servers, and spotted over 1,100, around 20 percent of which are "actively hosting models susceptible to unauthorized access."
Information security
[ Load more ]