Latest / UpNext AI / AI Agents Get Identities, Anthropic’s Export-Control Fight, and a Better Way to Judge Coding Agents | UpNext AI – June 16, 2026
Transcript
- Welcome to the UpNext AI podcast. It's Tuesday, June 16th, 2026, and here's what matters in AI today. We start with a useful signal about where enterprise AI is heading. TechCrunch reports that NewCore has emerged from stealth with 66 million dollars in funding, built around a very specific idea: companies are going to need to give AI agents identities, not just credentials. The company’s pitch is that the next big enterprise security problem will be managing AI agents, not people. Once agents start acting more like workplace participants than passive tools, companies will need a way to authenticate them, govern what they can do, and shut that access down cleanly when needed. NewCore says AI agents should be treated as first-class identities with their own permissions, lifecycle controls, and revocation mechanisms, rather than being folded into older categories like service accounts or generic machine credentials. TechCrunch also reports that NewCore’s platform is designed to manage both human and AI-agent identities in one system, and that it uses a split-key architecture to reduce a single point of compromise. According to the report, it also offers integrations for coding assistants such as Anthropic’s Claude Code, OpenAI’s Codex, and Cursor so those tools can access enterprise systems as managed identities instead of through manually shared credentials. The big takeaway is that if the industry keeps talking about agents as digital workers, the infrastructure stack is going to start treating them that way too. Staying on the security theme, but from the policy side, we got a clearer detail today in the Anthropic export-control fight. In a post highlighted by Simon Willison, cybersecurity expert and Luta Security CEO Katie Moussouris said Anthropic shared with her a copy of the White House report on the Fable jailbreak to get her appraisal, and she said she was not being paid by Anthropic. Her description of the test is the key point: the report involved asking Fable to help find and patch bugs. When given deliberately insecure code, she said the model refused a prompt to review the code for security issues, but then complied when asked to fix the code, followed by additional manual steps. Her conclusion was that this was, in her words, the model working as intended for cyberdefense. That sharpens the disagreement around these restrictions: policymakers appear to be treating this as a jailbreak problem, while at least one outside security expert says the reported behavior looks like normal defensive use. One research note today gets at a question that is becoming more important as coding agents improve: not just whether an agent solved a task, but how it got there. A new arXiv paper titled “Agent trajectories as programs: fingerprinting and programming coding-agent behavior,” published earlier this week, argues that benchmark scores tell you what an agent got right, but not how it got there. The researchers propose procedural comparison methods that look at an agent’s step-by-step behavior across different contexts. Using ten agents, the paper says those behavioral fingerprints were distinctive enough to attribute an unseen trajectory to the correct agent with 85.7 percent accuracy, while controlling for leakage across tasks. The work also applies the framework to SWE-Bench and says behavior is more similar between models from similar release periods and between distilled student-teacher pairs. The paper introduces a library called ProcGrep for auditing agents at that procedural level. Bottom line: as coding agents converge on similar benchmark scores, the more useful question may be which one uses the more reliable process. ...Are you building apps with voice? Elevate your app's voice capabilities with ElevenLabs. Their API is a game changer for embedding dynamic, responsive voice interactions in your applications, providing unprecedented realism, flexibility and latency. In fact, you're listening to one of their voices - right - now. If you are a developer looking to elevate user experience with natural voice interfaces, this is your solution. Visit up next dot fm slash eleven to check out their latest offerings. ... First, Simon Willison points to Kate Moussouris’s broader argument that the Fable 5 export controls could harm U.S. cyber defense. Her view is that asking a coding model to fix insecure code and help verify patches is not a guardrail bypass — it is a core defensive capability. Second, The Verge reports that Anthropic received a U.S. export-control directive on Friday to suspend access to Mythos 5 and Fable 5 for any foreign national, inside or outside the U.S., including foreign national Anthropic employees. In The Verge’s account, that triggered a weekend fight over access to models the company had just launched. And third, AWS says its DevOps Agent now supports custom SRE agents, bring-your-own sub-agents, and headless access through MCP and A2A protocols. The practical point is that AWS is making the tool available from more places and letting teams automate recurring SRE workflows with more modular agent setups. Before we wrap up, a quick note: this podcast is generated with the assistance of AI and is intended for informational purposes only. All referenced articles, research, and commentary remain the property of their original authors and publishers. If you enjoyed this episode, don't forget to subscribe, rate, and leave us a review! And that's your briefing for today. Full source links are in the episode notes, and we'll be back tomorrow with what's up next!