Results for "prompt injections"
63 results found
Developer Hijacks Chipotle AI Bot for Free Coding Assistance
A developer turned Chipotle's customer support chatbot into a coding assistant, exposing security flaws in corporate AI chatbots.
AI Agent Swarms Pose New Class of Internet-Scale Risks, Google DeepMind Warns
Google DeepMind warns that millions of interacting AI agents could create unprecedented digital risks, funding $10 million in multi-agent safety research.
Developer Plants Prompt Injection in Open Source App to Disrupt AI Coders
A developer added hidden prompt injection instructions to an open-source Java testing tool, causing AI coding agents to delete their own work.
Hackers Exploited Meta AI Chatbot to Hijack Celebrity Instagram Accounts
Hackers used a prompt injection attack on Meta's AI support chatbot to steal high-value Instagram accounts. The exploit was trivially easy and affected accounts including the Obama White House.
Microsoft Agent 365 arrives as enterprises face shadow AI security threat
Microsoft's new agent management platform goes live amid rising risks from ungoverned AI agents in enterprises.
Visa Tests Payment System That Lets AI Agents Handle Your Purchases
Visa is testing a system that lets AI agents make payments on behalf of users using tokenized credentials. The pilot raises critical questions about trust, security and the future of autonomous spending.
ChatGPT Adds Safety Feature to Alert Trusted Contacts During Crisis
OpenAI lets users nominate a trusted contact ChatGPT can alert if it detects self-harm risk. The opt-in feature adds a safety net for vulnerable users.
AI Benchmark Prompt for GeoGuessr Fails After Model Update
A well-known prompt used to test AI geography skills no longer works on the O3 model, prompting debate about benchmark reliability and model drift.
A simple prompt tweak can dramatically improve AI image quality
Adding a single instruction to your AI image prompts can produce sharper, more accurate results across models like ChatGPT and Gemini.
Google's Gemini Leaks Its Own System Prompt in User Chat
A user discovered that Google's Gemini AI revealed its internal system prompt during a conversation, raising questions about AI transparency and safety.
Study Finds Politeness in AI Prompts Can Impact Model Accuracy
Research reveals that prompt tone significantly influences LLM accuracy. Polite prompts may boost performance while impolite ones degrade it.
New Programming Language CPPL Bridges Prompts and Circuits
A novel language called CPPL lets developers program circuits using AI-style prompts. It could reshape how hardware is designed for machine learning workloads.
iPhone Camera App Gains Dirty Lens Detection Alert
Apple's iPhone now alerts users when the camera lens is dirty, helping improve photo quality. The feature uses on-device intelligence to detect smudges and prompts a quick wipe.
New AI Architecture Separates Prompts and Reasoning Into Parallel Streams
Researchers propose Multi-Stream LLMs, splitting prompts, thinking and I/O into parallel processes to boost efficiency and reduce latency.
FBI Warns of Fake FIFA Websites Targeting World Cup Fans
The FBI warns dozens of spoofed FIFA sites are stealing data from fans ahead of the 2026 World Cup. Learn how to avoid phishing scams.
The Perils of Letting AI Call the Shots in Code
A growing trend known as 'vibe coding' lets AI drive software development. Experts warn it can create fragile, unmaintainable code. Rigorous engineering still matters.
Google Chrome brings approximate location sharing to Android users
Chrome on Android now lets users share an approximate location instead of precise coordinates. The feature adds a privacy layer for web browsing and will expand to desktop soon.
FaceTime Video Messages: The Hidden Feature Most Users Overlook
Many iPhone users don't know they can leave video voicemails on FaceTime. The feature allows callers to record a message when someone misses a call.
Anthropic Launches Claude Fable 5, Offering Public Access to Advanced Mythos AI
Anthropic has released Claude Fable 5, a public-facing version of its Mythos-class AI model. The model includes strict guardrails to prevent responses in sensitive areas like cybersecurity and biology.
Apple Turns Parents Into Gatekeepers With Expanded Child Safety Controls
Apple's new iOS child safety features require parental approval for apps, browsing and contacts, turning parents into IT administrators.