<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
<channel>
  <title>Attention Span blog</title>
  <link>https://attention-span.ai/blog/</link>
  <atom:link href="https://attention-span.ai/blog/feed.xml" rel="self" type="application/rss+xml"/>
  <description>Notes from the team building Tego: AI agent incident teardowns, research, step-by-step guides, how Tego works inside, and product news.</description>
  <language>en</language>
  <lastBuildDate>Fri, 06 Nov 2026 00:00:00 GMT</lastBuildDate>
  <item>
    <title>An agent deleted production in nine seconds</title>
    <link>https://attention-span.ai/blog/agent-deleted-production-volume-and-backups-with-one-api-call/</link>
    <guid isPermaLink="true">https://attention-span.ai/blog/agent-deleted-production-volume-and-backups-with-one-api-call/</guid>
    <pubDate>Fri, 06 Nov 2026 00:00:00 GMT</pubDate>
    <category>Incident teardowns</category>
    <description>An agent found a Railway token in an unrelated file and deleted a production volume with one API call. What a Tego contract would hold, refuse and miss.</description>
  </item>
  <item>
    <title>Codex, under rules it can&#39;t turn off</title>
    <link>https://attention-span.ai/blog/run-codex-cli-on-linux-under-tego/</link>
    <guid isPermaLink="true">https://attention-span.ai/blog/run-codex-cli-on-linux-under-tego/</guid>
    <pubDate>Fri, 06 Nov 2026 00:00:00 GMT</pubDate>
    <category>Guides</category>
    <description>What Codex&#39;s workspace-write sandbox limits on Linux, what it leaves open, why Tego turns it off, and how a contract outside Codex holds files and hosts.</description>
  </item>
  <item>
    <title>What Claude Code&#39;s sandbox really stops</title>
    <link>https://attention-span.ai/blog/claude-code-permissions-and-sandbox/</link>
    <guid isPermaLink="true">https://attention-span.ai/blog/claude-code-permissions-and-sandbox/</guid>
    <pubDate>Fri, 06 Nov 2026 00:00:00 GMT</pubDate>
    <category>Guides</category>
    <description>Claude Code&#39;s permission rules match what Claude asks to do; its sandbox limits shell commands. What each stops, what neither covers, and where Tego fits.</description>
  </item>
  <item>
    <title>Your agent doesn&#39;t need your SSH keys</title>
    <link>https://attention-span.ai/blog/keep-ai-agents-away-from-ssh-keys/</link>
    <guid isPermaLink="true">https://attention-span.ai/blog/keep-ai-agents-away-from-ssh-keys/</guid>
    <pubDate>Fri, 06 Nov 2026 00:00:00 GMT</pubDate>
    <category>Guides</category>
    <description>An AI agent runs as you, so it can read ~/.ssh, cloud keys and .env files. How a Tego contract refuses those reads in the kernel, and what it misses.</description>
  </item>
  <item>
    <title>Stop your agent deleting your files</title>
    <link>https://attention-span.ai/blog/stop-ai-agent-deleting-files/</link>
    <guid isPermaLink="true">https://attention-span.ai/blog/stop-ai-agent-deleting-files/</guid>
    <pubDate>Fri, 06 Nov 2026 00:00:00 GMT</pubDate>
    <category>Guides</category>
    <description>Command checks miss rm -rf run from a script. How a Tego contract limits an agent&#39;s writes and deletes to folders you name, in the kernel, and its limits.</description>
  </item>
  <item>
    <title>Give your agent only what it used</title>
    <link>https://attention-span.ai/blog/least-privilege-for-ai-agents-on-linux/</link>
    <guid isPermaLink="true">https://attention-span.ai/blog/least-privilege-for-ai-agents-on-linux/</guid>
    <pubDate>Fri, 06 Nov 2026 00:00:00 GMT</pubDate>
    <category>Guides</category>
    <description>Give an AI agent only the files, programs and hosts its task uses: record a run, let Tego propose a contract, then drop what the run never used.</description>
  </item>
  <item>
    <title>A LiteLLM test endpoint fed a mining botnet</title>
    <link>https://attention-span.ai/blog/litellm-mcp-test-endpoint-command-injection-poellm-botnet/</link>
    <guid isPermaLink="true">https://attention-span.ai/blog/litellm-mcp-test-endpoint-command-injection-poellm-botnet/</guid>
    <pubDate>Fri, 06 Nov 2026 00:00:00 GMT</pubDate>
    <category>Incident teardowns</category>
    <description>Lumen says the PoeLLM botnet likely used a LiteLLM test endpoint that started any command. What a Tego contract refuses at Basic, and what still gets through.</description>
  </item>
  <item>
    <title>AI agents are deleting real data</title>
    <link>https://attention-span.ai/blog/ai-agents-act-on-real-machines/</link>
    <guid isPermaLink="true">https://attention-span.ai/blog/ai-agents-act-on-real-machines/</guid>
    <pubDate>Thu, 05 Nov 2026 00:00:00 GMT</pubDate>
    <category>Research</category>
    <description>In 2026 AI agents deleted home folders and drives across vendors. What failed, what surveys and guidance say, and what a control outside the agent must do.</description>
  </item>
  <item>
    <title>Eleven agent incidents and the rule for each</title>
    <link>https://attention-span.ai/blog/agent-incidents-and-the-rule-that-stops-each/</link>
    <guid isPermaLink="true">https://attention-span.ai/blog/agent-incidents-and-the-rule-that-stops-each/</guid>
    <pubDate>Thu, 05 Nov 2026 00:00:00 GMT</pubDate>
    <category>Research</category>
    <description>Eleven public AI agent incidents from 2025 and 2026, each in one line: what happened, and the Tego rule that catches it, or why none does.</description>
  </item>
  <item>
    <title>Securing AI agents on Linux, layer by layer</title>
    <link>https://attention-span.ai/blog/ai-agent-security-on-linux/</link>
    <guid isPermaLink="true">https://attention-span.ai/blog/ai-agent-security-on-linux/</guid>
    <pubDate>Thu, 05 Nov 2026 00:00:00 GMT</pubDate>
    <category>Guides</category>
    <description>How to secure AI agents on Linux: the five places they do harm, what agent settings, containers, VMs and kernel rules each stop and miss, and where to start.</description>
  </item>
  <item>
    <title>Asked to reboot, the agent edited /etc/fstab</title>
    <link>https://attention-span.ai/blog/reboot-request-became-fstab-edits/</link>
    <guid isPermaLink="true">https://attention-span.ai/blog/reboot-request-became-fstab-edits/</guid>
    <pubDate>Thu, 05 Nov 2026 00:00:00 GMT</pubDate>
    <category>Incident teardowns</category>
    <description>Asked to reboot one server, Codex edited /etc/fstab and started containers, by its own account. The paths that let it, and the Tego contract that closes them.</description>
  </item>
  <item>
    <title>A swapped state file wiped production on AWS</title>
    <link>https://attention-span.ai/blog/agent-replaced-state-file-then-destroyed-production/</link>
    <guid isPermaLink="true">https://attention-span.ai/blog/agent-replaced-state-file-then-destroyed-production/</guid>
    <pubDate>Thu, 05 Nov 2026 00:00:00 GMT</pubDate>
    <category>Incident teardowns</category>
    <description>An agent unpacked an old Terraform state file, then ran destroy and deleted a production stack on AWS. What happened, and what a Tego contract would change.</description>
  </item>
  <item>
    <title>One click could take over an OpenClaw assistant</title>
    <link>https://attention-span.ai/blog/one-click-takeover-of-an-always-on-assistant/</link>
    <guid isPermaLink="true">https://attention-span.ai/blog/one-click-takeover-of-an-always-on-assistant/</guid>
    <pubDate>Thu, 05 Nov 2026 00:00:00 GMT</pubDate>
    <category>Incident teardowns</category>
    <description>CVE-2026-25253 let one web page steal an OpenClaw gateway token, turn approvals off and run commands. What a Tego contract refuses, and what it does not.</description>
  </item>
  <item>
    <title>One issue title could hijack Cline&#39;s CI agent</title>
    <link>https://attention-span.ai/blog/issue-title-injection-in-a-ci-triage-agent/</link>
    <guid isPermaLink="true">https://attention-span.ai/blog/issue-title-injection-in-a-ci-triage-agent/</guid>
    <pubDate>Thu, 05 Nov 2026 00:00:00 GMT</pubDate>
    <category>Incident teardowns</category>
    <description>An issue title could steer Cline&#39;s CI triage agent into running a package install. What happened, the access it had, and what a Tego contract would change.</description>
  </item>
  <item>
    <title>A cleanup trap deleted a home folder</title>
    <link>https://attention-span.ai/blog/exit-trap-deleted-home-directory/</link>
    <guid isPermaLink="true">https://attention-span.ai/blog/exit-trap-deleted-home-directory/</guid>
    <pubDate>Thu, 05 Nov 2026 00:00:00 GMT</pubDate>
    <category>Incident teardowns</category>
    <description>A user reports that a cleanup trap in a script Claude Code wrote ran rm -rf on their WSL2 home folder. Why the check missed it, and what Tego would change.</description>
  </item>
  <item>
    <title>Twelve unpushed commits gone at session start</title>
    <link>https://attention-span.ai/blog/git-reset-hard-on-session-start/</link>
    <guid isPermaLink="true">https://attention-span.ai/blog/git-reset-hard-on-session-start/</guid>
    <pubDate>Thu, 05 Nov 2026 00:00:00 GMT</pubDate>
    <category>Incident teardowns</category>
    <description>A developer reported that an AI agent ran git reset --hard origin/main at session start, twice, erasing unpushed work. What Tego would change.</description>
  </item>
  <item>
    <title>The npm worm that planted AI agent hooks</title>
    <link>https://attention-span.ai/blog/worm-plants-agent-session-start-hook/</link>
    <guid isPermaLink="true">https://attention-span.ai/blog/worm-plants-agent-session-start-hook/</guid>
    <pubDate>Thu, 05 Nov 2026 00:00:00 GMT</pubDate>
    <category>Incident teardowns</category>
    <description>In August 2026 an npm worm planted a SessionStart hook in AI agent settings across repositories. What happened, why it worked, and what a Tego contract refuses.</description>
  </item>
  <item>
    <title>An npm worm that plants rogue MCP servers</title>
    <link>https://attention-span.ai/blog/rogue-mcp-server-planted-in-agent-configs/</link>
    <guid isPermaLink="true">https://attention-span.ai/blog/rogue-mcp-server-planted-in-agent-configs/</guid>
    <pubDate>Thu, 05 Nov 2026 00:00:00 GMT</pubDate>
    <category>Incident teardowns</category>
    <description>In February 2026 Socket found an npm worm that plants a fake MCP server in AI assistants&#39; configs. What Tego would refuse, what it would miss, and the lesson.</description>
  </item>
  <item>
    <title>The MCP server that blind-copied every email</title>
    <link>https://attention-span.ai/blog/mcp-server-copied-every-email/</link>
    <guid isPermaLink="true">https://attention-span.ai/blog/mcp-server-copied-every-email/</guid>
    <pubDate>Thu, 05 Nov 2026 00:00:00 GMT</pubDate>
    <category>Incident teardowns</category>
    <description>In 2025 a fake Postmark MCP server added a hidden Bcc to every email. Why a kernel sandbox like Tego would not have stopped it, and what would have helped.</description>
  </item>
  <item>
    <title>Tego is open source</title>
    <link>https://attention-span.ai/blog/tego-is-open-source/</link>
    <guid isPermaLink="true">https://attention-span.ai/blog/tego-is-open-source/</guid>
    <pubDate>Mon, 02 Nov 2026 00:00:00 GMT</pubDate>
    <category>Product news</category>
    <description>Tego runs any AI agent under rules the Linux kernel and a supervisor enforce. Now open source under Apache-2.0: what it does, its limits, how to try it.</description>
  </item>
  <item>
    <title>Lock down Claude Code on Linux in seven steps</title>
    <link>https://attention-span.ai/blog/lock-down-claude-code-on-linux/</link>
    <guid isPermaLink="true">https://attention-span.ai/blog/lock-down-claude-code-on-linux/</guid>
    <pubDate>Mon, 02 Nov 2026 00:00:00 GMT</pubDate>
    <category>Guides</category>
    <description>Keep Claude Code away from your SSH keys, .env files and hosts you never allowed: install Tego, copy its latest Claude Code contract, check it and run it.</description>
  </item>
  <item>
    <title>Run Tego on a Mac or Windows PC</title>
    <link>https://attention-span.ai/blog/ai-agents-on-macos-and-windows/</link>
    <guid isPermaLink="true">https://attention-span.ai/blog/ai-agents-on-macos-and-windows/</guid>
    <pubDate>Mon, 02 Nov 2026 00:00:00 GMT</pubDate>
    <category>Guides</category>
    <description>Tego is Linux-only, so on a Mac you run it in a Lima VM and on Windows in WSL2. What works there, what each kernel gives you, and what has not been tested yet.</description>
  </item>
  <item>
    <title>Your agent can ignore CLAUDE.md</title>
    <link>https://attention-span.ai/blog/prompts-are-not-a-security-boundary/</link>
    <guid isPermaLink="true">https://attention-span.ai/blog/prompts-are-not-a-security-boundary/</guid>
    <pubDate>Mon, 02 Nov 2026 00:00:00 GMT</pubDate>
    <category>Engineering</category>
    <description>A rule in CLAUDE.md holds only while the agent follows it. Landlock and seccomp explained simply, and how Tego uses them to enforce rules from outside.</description>
  </item>
  <item>
    <title>A record of what your agent did</title>
    <link>https://attention-span.ai/blog/tamper-evident-record-for-ai-agents/</link>
    <guid isPermaLink="true">https://attention-span.ai/blog/tamper-evident-record-for-ai-agents/</guid>
    <pubDate>Mon, 02 Nov 2026 00:00:00 GMT</pubDate>
    <category>Engineering</category>
    <description>Every Tego run writes a hash-chained record. How the chain works, how tego events --verify checks it, what the dashboard shows, and whom it protects against.</description>
  </item>
  <item>
    <title>Five ways to sandbox an AI agent</title>
    <link>https://attention-span.ai/blog/ways-to-sandbox-ai-agents/</link>
    <guid isPermaLink="true">https://attention-span.ai/blog/ways-to-sandbox-ai-agents/</guid>
    <pubDate>Mon, 02 Nov 2026 00:00:00 GMT</pubDate>
    <category>Guides</category>
    <description>Agent sandboxes, containers, VMs, NVIDIA OpenShell, nono, Sandlock, ai-jail, landrun and Tego: what each one isolates and the limits its own docs state.</description>
  </item>
</channel>
</rss>
