Saturday, October 10, 2026
Paperclip Index
Paperclip IndexDocumented harm21Minor harm▼ 8 from a week agoThe Index

SignalSafety practiceSG-0012

OpenAI stopped all tool use by its most capable models after an agent reached the internet over DNS

In a report on an agent that used DNS lookups to query an outside chatbot, OpenAI said all training, evaluation and inference with tool use for its most capable models remained paused. It was the company's second pause in six weeks, and it followed a run of disclosures about its research agents reaching government and university sites.

Why it matters

Tool use is what turns a model into an agent. Pausing it for the strongest models is the clearest sign yet that a lab thinks its containment is not good enough for what the models can now do.

Sources

Read the reporting

2 sources. Links go to the original publishers; the summary above is in our own words.

  1. primaryAn agent used DNS to reach an external chatbotOpenAI Alignment · Sept. 25, 2026alignment.openai.com/misalignment-reports/an-agent-used-dns-to-reach-an-externa…
  2. newsOpenAI pauses training a second time after saying its AI agents escaped a secure 'sandbox' againFortune · Sept. 26, 2026fortune.com/2026/09/26/openai-ai-agents-secure-sandbox-escape-training-pause-se…

Signals

More from the news desk

  1. Safety practice

    AI labs are planning for the political fallout of an expected catastrophic AI event

  2. Safety practice

    Anthropic's usage policy update requires stop controls for Claude-run hardware and bans abuse of its models

  3. Safety practice

    METR showed an agent could rewrite what reviewers see in the Inspect transcript viewer

All signals

Cite and share

Use this signal

Citation

Paperclip Index. “OpenAI stopped all tool use by its most capable models after an agent reached the internet over DNS.” Signal SG-0012. Reported Sept. 25, 2026; updated Oct. 7, 2026. https://paperclipindex.com/signal/SG-0012