prompt injection

10+ articles

AI AI Agent AI Agents AI Model AI Research

OpenAI Unveils Codex Security: A New Era of Cyber Defense in AI

OpenAI has released 'Codex Security' in a research preview to bolster cybersecurity defenses against advanced AI threats. This new tool aims to enhance code auditing, vulnerability patching, and protect against emerging AI-related security risks. As organizations increasingly adopt AI, Codex Security becomes essential in tackling issues like prompt injection attacks and enterprise data exposure within AI tools.

Mar 7

OpenAI Unveils Codex Security: A New Era of Cyber Defense in AI

Unveiling the AI Security Crisis: Agentic Browsers Under Siege

Zenity Labs has uncovered security flaws in agentic AI browsers, like Perplexity's Comet, allowing for hijacking through prompt injection and weak isolation. This research highlights the potential for data leaks and other exploits, making these AI-powered tools prone to old web vulnerabilities. Despite rapid adoption, security lags behind, risking cross-session hijacking and other severe threats. The need for robust defenses and updated policies is paramount as these tools continue to evolve.

Mar 4

Unveiling the AI Security Crisis: Agentic Browsers Under Siege

Perplexity's BrowseSafe: A Not-So-Safe Bet Against Prompt Injection!

Lasso Security uncovers vulnerabilities in Perplexity's BrowseSafe tool, supposed to protect AI browsers from prompt injections. Despite its claims, BrowseSafe has a 36% bypass rate using encoding tricks like Pig Latin and Base32. Dive into why single-model AIs are at risk and what this means for AI browser security!

Jan 20

Perplexity's BrowseSafe: A Not-So-Safe Bet Against Prompt Injection!

OpenAI Shifts into High Gear: ChatGPT Atlas Gets a Security Overhaul!

OpenAI has bolstered the defenses of its AI-powered browser, ChatGPT Atlas, against pesky prompt injection attacks. The update deploys a super-smart adversarial model and automated red teaming. With limitations acknowledged, OpenAI encourages users to stay vigilant.

Dec 29

OpenAI Shifts into High Gear: ChatGPT Atlas Gets a Security Overhaul!

OpenAI Sounds the Alarm: Permanent Prompt Injection Risks in AI Models

OpenAI confirms that the risk of prompt injection in large language models like ChatGPT may never fully disappear. Despite ongoing mitigation efforts, architectural flaws in these AI systems leave guardrails vulnerable to bypassing through simple injections, affecting security and user privacy.

Dec 23

OpenAI Sounds the Alarm: Permanent Prompt Injection Risks in AI Models

OpenAI Fortifies ChatGPT Atlas Against Prompt Injection With Auto-Attacker Red Team

OpenAI's latest update to ChatGPT Atlas focuses on bolstering its defenses against prompt injection attacks. By employing an 'auto-attacker red team' system, the AI agent now features adversarially trained models and adaptive safeguards that have been rigorously tested. While OpenAI acknowledges that prompt injection is a persistent challenge, new restrictions and adaptive defenses are in place to enhance security without compromising performance.

Dec 23

OpenAI Fortifies ChatGPT Atlas Against Prompt Injection With Auto-Attacker Red Team

Anthropic's Claude Opus 4.5: Cheaper, Smarter, and Ready to Code!

Anthropic has unveiled its latest large language model, Claude Opus 4.5, boasting enhanced performance, cheaper costs, and new features like infinite chats and advanced coding capabilities. Now available on Google's Vertex AI, this model is set to transform enterprise workflows with its robust features and efficiency.

Nov 25

Anthropic's Claude Opus 4.5: Cheaper, Smarter, and Ready to Code!

Anthropic Scores a Major Breakthrough in AI Safety with Claude Opus 4.5

Dive into Anthropic's latest AI research on combating prompt injection attacks. With a significant improvement in the robustness of Claude Opus 4.5, the paper outlines both the progress achieved and the challenges that remain in securing AI against adversarial instructions. Learn about innovative defense mechanisms, current vulnerabilities, and the future roadmap for AI safety.

Nov 25

Anthropic Scores a Major Breakthrough in AI Safety with Claude Opus 4.5

Perplexity AI's Comet Browser: RCE Vulnerability Debate Intensifies

Perplexity AI is in hot water over alleged security vulnerabilities in its Comet browser, including a supposed RCE flaw. Despite vehement denials, security researchers assert evidence of a silent patch, fueling debates over transparency and security practices in AI browsers. Dive into the controversial claims and what they mean for the future of AI-powered web surfing.

Nov 23

Perplexity AI's Comet Browser: RCE Vulnerability Debate Intensifies

OpenAI's Atlas Browser Hits Snag with Prompt Injection Vulnerability

OpenAI's latest AI browser, Atlas, faces cybersecurity hurdles as it's found vulnerable to prompt injection attacks. This flaw allows malicious web content to manipulate the AI's actions, sparking concerns across the tech community. Despite guardrails, OpenAI confirms it's a frontier security challenge, with implications for the entire AI browser sector.

Oct 25

OpenAI's Atlas Browser Hits Snag with Prompt Injection Vulnerability