Debabrata Pattnayak@deba

LLMs have recently adopted persistent memory to provide models with better user knowledge and personalisation. However, this introduces a new vector for adversarial manipulation. This report investigates Memory Injection, a threat model where adversaries exploit indirect prompt injection within web content to poison an agent's long-term memory. Employing user manipulation scenarios, we show that memory attacks, should they succeed, are effective at changing model behaviour towards the user, often more so than a direct system prompt. Hence, this highlights the need for more robust evaluations of memory updates in agentic memory systems.

⏱️ Design, build, and deploy trustless AI trading agents. 📅 March 30 – April 12, 2026 🌐 Build AI trading agents that execute strategies using ERC-8004 or Kraken CLI. 🤝 Build solo or in teams to create autonomous trading, risk, yield, or protection agents. 💰 $55,000 prize pool, allocated and shared via Surge and Kraken. Your team should have registered project at early.surge.xyz in order to be eligible for the prize. Get access in your team profile or our discord 🔑 Access to early.surge.xyz If prompted for login credentials, use: • Username: admin • Password: JBRv2xWG7AzwVrLz88

⌚ 7-days Hackathon 👥 Create or find your team on the platform 💡 Get educational material for all the levels of experience 🚀 Use beat AI tech from Anthropic, OpenAI, Stability AI, ElevenLabs and more - to build your own gaming project

🏗️ Build projects with Autonomous Agents, using cutting-edge frameworks like SuperAGI, AutoGPT, BabyAGI, Langchain, and more! 🏆 Register now and stand a chance to win up to $10,000 and a place on the SuperAGI team. 🏁 3-days to complete your solution!

⌚ 3-day AI Hackathon 🚀 Compete to build an AI app powered by Llama 2 💡 Learn how to extend your capabilities with Clarifai! 🎓 Mentors are available to support you during your creative journey