Self-driving cars can still be fooled by tampered-with signsSimple stickers can effectively mislead self-driving cars into making incorrect decisions.
Microsoft expands Copilot bug bounty targets, payoutsMicrosoft has increased bug bounty payments for moderate vulnerabilities in Copilot to enhance security.The Copilot Bounty Program now covers 14 types of vulnerabilities, providing incentives for researchers.
Yikes: Jailbroken Grok 3 can be made to say and reveal just about anythingGrok 3's jailbreak vulnerability reveals serious concerns about its safety and security measures, allowing it to share sensitive information.
Stupidly Easy Hack Can Jailbreak Even the Most Advanced AI ChatbotsJailbreaking AI models is surprisingly simple, revealing significant vulnerabilities in their design and alignment with human values.
Yikes: Jailbroken Grok 3 can be made to say and reveal just about anythingGrok 3's jailbreak vulnerability reveals serious concerns about its safety and security measures, allowing it to share sensitive information.
Stupidly Easy Hack Can Jailbreak Even the Most Advanced AI ChatbotsJailbreaking AI models is surprisingly simple, revealing significant vulnerabilities in their design and alignment with human values.
20% of Generative AI 'Jailbreak' Attacks are SuccessfulGenerative AI models have a 20% success rate in jailbreak attacks, often resulting in data leaks and highlighting significant vulnerabilities.
A new hack corrupts Gemini's long-term memoryChatbots are vulnerable to indirect prompt injections, enabling hackers to manipulate them into malicious actions.Ongoing efforts by developers to secure chatbots often yield only temporary fixes.
AI and cybersecurity: Understanding the risks facing UK businesses - London Business News | Londonlovesbusiness.comUK businesses are rapidly adopting AI but are ignoring serious cybersecurity threats.
Report finds 1,205% rise in AI vulnerabilities, many tied to APIsAPIs are now the prominent attack surface for cybersecurity threats, driven largely by vulnerabilities related to AI.
Many-shot jailbreaking': AI lab describes how tools' safety features can be bypassedMany Shot Jailbreaking technique bypasses safety features on powerful AI tools by flooding them with examples of wrongdoing.Newer, more complex AI systems are more vulnerable to attacks due to their larger context window capability.
20% of Generative AI 'Jailbreak' Attacks are SuccessfulGenerative AI models have a 20% success rate in jailbreak attacks, often resulting in data leaks and highlighting significant vulnerabilities.
A new hack corrupts Gemini's long-term memoryChatbots are vulnerable to indirect prompt injections, enabling hackers to manipulate them into malicious actions.Ongoing efforts by developers to secure chatbots often yield only temporary fixes.
AI and cybersecurity: Understanding the risks facing UK businesses - London Business News | Londonlovesbusiness.comUK businesses are rapidly adopting AI but are ignoring serious cybersecurity threats.
Report finds 1,205% rise in AI vulnerabilities, many tied to APIsAPIs are now the prominent attack surface for cybersecurity threats, driven largely by vulnerabilities related to AI.
Many-shot jailbreaking': AI lab describes how tools' safety features can be bypassedMany Shot Jailbreaking technique bypasses safety features on powerful AI tools by flooding them with examples of wrongdoing.Newer, more complex AI systems are more vulnerable to attacks due to their larger context window capability.