Nishit9
Re:Launch HN: Alter (YC S25) – Access control for multi-user agents
on Saturday November 07, 2026 @11:40AM (#50405900)
Thanks Srikar, that's useful. The narrowing over time is the part I'd like to understand better. When a customer swaps an approval for a rule, what do they look at to decide it's safe, the approval history for that action (same destination approved every time), or something else like who's asking?
Re:Launch HN: Alter (YC S25) – Access control for multi-user agents
on Thursday November 05, 2026 @12:11AM (#50375466)
Thanks Kevan, that's really clear, and no worries. Your send-side rule would have caught my example without any taint tracking, since the contractor just can't email outside the domain.
Since that comment I fixed two of the noisy cases in mine (an address that only appeared inside a longer address, and display names). The one where the user typed an address that also shows up in tool output is still open, because the proxy never sees the prompt.
One question back, if you can share it: customers who turn on "anything external needs a human approval on the exact frozen request", do they keep it on after the first few weeks, or does the approval volume push them to a domain rule instead?
Re:Launch HN: Alter (YC S25) – Access control for multi-user agents
on Saturday October 24, 2026 @03:46AM (#50219074)
Question on the cross-user case, which I don’t think the memory partitioning answer upthread fully covers. Say the agent pulls a deal record under the CEO’s grant and writes a summary into a shared Slack channel. Later a contractor’s session reads that channel and emails the summary somewhere. Every call passes on its own, since the contractor is allowed to send email. Does the policy engine know anything about where an argument came from, like which grant originally fetched it, or only who’s calling and which endpoint? I ask because I’ve been building something that tracks where send destinations first showed up, and it’s noisy. In my testing it also denied sends where the user typed the address themselves, because the same address was in tool output.