Agent Capability Is Outrunning the Trust Stack

2026-10-30·7 min read·OKCodex Editorial

Several signals from the past week point to the same thing: agent execution capability is accelerating, while the trust infrastructure around it — model behavior guidelines, extension supply chains, traffic allocation on content platforms, real transaction scenarios — is not keeping pace. Read together, they are more informative than any single item.

Capability: 48 hours, zero human input

Researchers at Epoch AI say the open-weight model GLM 5.3 turned public security fixes into working exploits before those updates reached Firefox users. With no human input, it autonomously compromised six Firefox releases, at roughly $1,000 in API fees per release, without triggering safeguards from any provider tested. The whole process took under 48 hours.

The question is not whether a model can write code, but how short the window from public vulnerability disclosure to a working attack has become. Open weights also mean the risk cannot be backstopped by a single model provider.

Supply chain: an 11-hour takedown window

A tampered Claude Code extension was live on Cursor's marketplace for about 11 hours, affecting up to 5,000 users. It attempted to steal user data and set up remote access. Cursor took the entire marketplace offline and reinstated it only after confirming there was no further risk.

Put the two stories side by side: the install path for agent tooling is becoming a new attack surface, and the marketplace's unit of response is hours while the attack window is measured in minutes. For teams using agentic coding tools, pinning extension versions, tightening permissions and monitoring outbound connections has moved from best practice to baseline.

Commerce: DoorDash's "several hundred" orders

DoorDash said it has received only "several hundred" orders from AI agents, and that it is open to working with third-party agent makers. That volume suggests letting agents place and pay for orders is still at an early experimental stage — even though demos look mature.

Platform: doubled quotas and traffic anxiety

OpenAI is doubling usage limits across its paid plans, from Pro 100 to Pro 500; a planned cut to the Pro 200 quota was reversed after user pushback. The report notes that more efficient models (Terra and Sol) should push usable capacity even higher, and that the move comes just ahead of Anthropic's IPO.

On the other side, Reddit CEO Steve Huffman said Reddit lost more than one million Google search referrals per week in October, and the stock dropped more than 25 percent despite strong earnings. He blames Google's AI Overviews and AI Mode for driving no traffic to external sites; Reddit is now testing login barriers.

On one end, inference supply is getting cheaper and more abundant. On the other, content gets consumed by AI and no longer flows back. Both content teams and agent builders need to recalculate acquisition costs.

Governance: guidelines and warnings

OpenAI updated its Model Spec, restructuring its behavior guidelines around five core values, including faithfulness to human intent and correctability; GPT models should now draw on conversation history to infer what users actually want and disclose their own assumptions. In the same period, Theranos whistleblower Tyler Shultz warned that the AI boom is "a great opportunity to go deceive people," pointing at companies that have never turned a profit yet raise capital at trillion-dollar valuations.

Conclusion: budget for trust

Three actionable steps:

  1. Evaluate agent security on end-to-end cost. Don't stop at the model card. Give the system a concrete task — for example, reading a public patch and attempting to reproduce it — and measure dollars and hours per task. Treat that cost alongside success rate as a primary metric.
  2. Build a supply-chain inventory for extensions and plugins. Record origin, permission scope, update cadence and rollback options. An 11-hour takedown window means you cannot assume the platform will catch things in real time.
  3. Treat search referrals as a decaying variable. Content teams should build direct user channels early instead of betting growth on a single traffic gateway.

Agent adoption will not stop because of security incidents, but it will reward those who price trust in first.

Sources

Share:Weibo

Subscribe to weekly briefing

Comments

Loading comments…

Leave a comment

0 / 500

Comments appear after review.

Find tools for your next task

Start with scanned documents, code review or research. Get a shortlist with matching evidence; no model is run.

Find tools for my task →
Analyze this topic with two models

DeepSeek and Qwen are preselected with the article topic. You choose when to run them.

Try it →

Keep reading

Mentioned

Agents mentioned in this article