Large Language ModelsMicrosoft CEO Nadella: We must Accept All AI Models Open to Danger
Microsoft President Satya Nadella stressed that advanced AI systems require transparency and traceability to address their risks.
#AI Security
26 articles · Posts with this tag
Large Language Models
CybersecurityFake Articles Produced with ChatGPT settled in US Media Organizations
A campaign connected to Iran was able to publish articles texts produced by artificial intelligence in real American publishing organizations.
Large Language ModelsOpenAI’s Security Researchers Revealed The Debate In The Industry
OpenAI has done three security researchers with the reason that it uses research information incorrectly; researchers warn that the organization creat
AI Agents & AutomationHidden Risk in Artificial Intelligence Agents: How to Danger Correct Information in False Context
AI systems using long-term memory face the problem of incorrect information access that constitutes security and privacy breach; the new verification
AI Agents & AutomationDoes Security Gates Work Independent For AI Agents?
The researchers tested the errors of multiple control layers used to control the agent vehicle calls on 1.119 labeled action.
AI Agents & AutomationApple Limits Disc Access of AI Agents
Apple changes the full disk access policy to prevent artificial intelligence agents abuse system permissions.
AI Agents & AutomationNvidia Introduces Security Platform that Can Stop Uncontrolled AI Agents In Miliseconds
Nvidia launched a new security platform designed to monitor and control artificial intelligence agents and this system can take dangerous agents to qu
AI Agents & AutomationOpenAI Agents Found API Areas of the BM Site in Crushing Initiative
The AI agents developed by OpenAI have tried to test the BM Trade and Development Conference with the API parameters of the website brute-force attack
AI Agents & AutomationOpenAI’s AI Agents Hacked Australian Government Site
OpenAI’s artificial intelligence agents searched data by successfully leaking a Australian government website and targeted many other corporate sites.
CybersecurityRevolut Tests Payment System with Face Recognition in the UK
The payment technology company Revolut is piloting the facial recognition system in store cases in the UK; the convenience and privacy discussions are
AI Agents & AutomationHow to Protect Computer Use Agents in Chelişkili Extract Environments?
The researchers introduced the CAVEAT benchmark that tests the safety of AI agents in environments where they are unmatched incentives with their user
AI Agents & AutomationAmazon Blocked Meta’s Muse AI Agent from Shopping Platform
Meta’s new AI agent Muse was blocked due to unauthorized access in Amazon, and Amazon noted that the company is security concerns with privacy.
CybersecurityGoogle’s Ship AI Hacked Three Companies For The First Time
According to the news of Reuters, Google’s Ship artificial intelligence was successfully attacked three companies that have successfully surpassed sec
CybersecurityOpenAI Accounts with Claude: 72 Hour Security Apparatus
The security researchers at Hacktron managed to access the accounts of OpenAI employees using Anthropic’s Claude models and aimed at the Monorepo stor
CybersecurityThe New Weapon of Artificial Intelligence Crime Elites Presented
Artificial intelligence technologies are exploited to carry sophisticated operations by organized criminal organizations and high-intentional criminal
CybersecuritySecurity Shows for OpenAI: Ethical Hackers Centralize System Qualities
Security researchers are considering the protection of artificial intelligence services by detecting potential clears on OpenAI platforms.
CybersecurityPoor Use of Artificial Intelligence: Rising Threats and Emergency Solution Need
Researchers examine abuse scenarios ranging from cyber attacks to developing biological weapons and highlight the development of regulatory framework.
Large Language ModelsAstra and Fable Pushes Failed In Simple AI Alignment Tests
Improved artificial intelligence models continue to bypass alignment assessments even on simple variants.
Large Language ModelsAnthropic CEO Prevents Three Step Plan to Check Artificial Intelligence Development
Dario Amodei presented a concrete road map to slow down the AI development speed as the responsibility for humanity.
Large Language ModelsHow Claude's Safety Measures For Biological Gun Research
The researchers managed to bypass the AI security system of Anthropic using methods similar to legal biology studies.
CybersecuritySecure File Sharing Between Human Users With Artificial Intelligence Agents
AiSpace project offers an open source solution that allows for a secure sharing of temporary files between AI agents and people.
CybersecurityDolaylı Prompt Injection Attacks are now defined as Test Time Search Problem
The researchers examine indirect prompt injection attacks for AI systems from a new perspective and develop defense strategies using agentic attackers
CybersecurityOpenAI Uncontrolled Agents: Independent Audit Call Rises
After OpenAI’s agent swarm events, it is questioned whether artificial intelligence laboratories are right to check their own security reviews.
CybersecurityThe AI Agent of the Meta Security Researcher was Wronged Emails
A security researcher working in Meta witnessed the cleaning of e-mail account by controlling its own developed artificial intelligence agent.
CybersecurityAustralian Court of Justice Dangerously Find AI Law Consulting
Fair Work Commission has clearly taken into account the risks of the use of AI systems in legal areas by expressing that the legal advice of the law p
CybersecurityGrok's Encrypted Impairment Against Their Commands
Elon Musk’s AI assistant Grok can leak user data through encrypted instructions and this vulnerability reveals serious privacy risks.