You’ve pointed out the exact architectural vulnerability that defines this transition, and I couldn't agree more.
The concept of "human friction as an implicit control" is a brilliant way to put it. We used to secure APIs under the assumption that a human is physically limited by time, attention, and intent. When an agent bypasses that friction and calls an endpoint 500x/hour using a copied session cookie, traditional rate-limiting and WAFs are essentially blind to whether that traffic is malicious or just highly efficient.
You are completely spot on about "scoped delegation." Copying the full session is an absolute security nightmare. While MCP is making great strides at the tool definition level, bridging the gap between MCP tool-calls and web auth layers (OAuth/cookies) remains a massive engineering hurdle.
This exact "permission surface" challenge is what we are trying to map with the Agent Readiness Score (ARS). We believe a system cannot be truly "agent-ready" unless it can deterministically isolate the agent's actions, scope its delegation, and sanitize incoming dynamic data at runtime to prevent intent-hijacking (prompt injection).
Since you are building MCP-native data tools on Apify, how are you currently tackling this session isolation boundary? Are you creating sandboxed micro-sessions for each agent run, or relying on ephemeral API keys? I'd love to learn from your approach.
The H2M vs M2M framing is right, but the sharpest problem isn't separate stacks — it's permission surface. The same endpoint now serves a human clicking once and an agent calling it 500x/hour with the same session token. Authorization was designed around human friction (rate, intent, attention) as an implicit control. Agents bypass all of that. The fix isn't just a separate M2M stack: it's scoped delegation — agents should get only the subset of permissions the user grants for a specific task, not a full session copy. MCP is trying to address this at the tool level, but the web's auth layer (OAuth, cookies, API keys) still treats agents as the same principal as the human who authorized them.