It’s now obvious that memory is one of the biggest bottlenecks in AI But the next question is whether that memory can be trusted “Context is the new bottleneck” - Jensen Huang at CES 2026 He’s right, but only half right Every AI agent deployed today still has the memory of a goldfish because context windows are stateless by design, starting each session from zero That is unfortunately how LLMs work But storage capacity is only half the battle, the other half is trust Current memory providers can write memories to docs, databases or internal logs, then ask enterprises to trust that record later The problem is obvious: one altered memory corrupts every decision that relied on it As agents move into regulated workflows, this becomes a real issue The EU AI Act requires high-risk AI systems to maintain automatic event logging and traceability, meaning enterprises will need proof of what data an agent accessed, what it remembered and whether that record was tampered with Our latest report covers AI memory and how Walrus tackles this problem with portable, verifiable and programmable memory for agents The full report is in the next post below