Social Engineering Won’t End; It’s Shifting to Your Agents

Social engineering isn’t about to end. The real shift is who gets targeted. As LLMs become embedded in operating systems and identity systems, attackers will pivot from exploiting human trust to manipulating AI agents—the new weakest link.

The conversation reveals a deeper problem: agentic identity. Estonia’s proposal for AI agents to get personal IDs hints at future IAM, but also amplifies the risk of authenticated malicious agents. More agents means more attack surface, not less, unless identity architectures evolve.

Engineering leaders should watch how authentication moves from static credentials to behavioral patterns. The idea of an LLM learning your habits to verify you is promising, but randomness and false positives remain unsolved. You can’t steal a pattern, but you can trick the model that reads it.

The immediate takeaway: governance for non-human identities is now a red-team priority. The IBM–OpenAI partnership on appsec is noise unless your team is a direct consumer. Focus on securing agent-to-agent interactions before AI agents make trust decisions in production.