Import AI 441: My agents are working. Are yours?
Practitioner-level accounts of agentic AI and an emerging data-poisoning threat together signal real near-term operational and integrity risks for APS AI deployments.
Key points
- An Anthropic researcher describes daily use of AI agents completing multi-day research tasks autonomously while he sleeps or hikes.
- Anti-AI activists have released 'Poison Fountain', a tool designed to corrupt AI training data via web crawlers.
- Eric Drexler's new paper frames AI governance around institutions directing many AI services, not singular systems.
Summary
This edition of Import AI covers three threads: a reflective essay by Anthropic's Jack Clark on his routine use of AI agents to complete research tasks autonomously; the emergence of 'Poison Fountain', an activist tool designed to corrupt AI training data by feeding poisoned content to web crawlers; and a summary of Eric Drexler's new framework arguing that AI governance should focus on institutions shaping many AI services rather than managing singular superintelligent agents. Together, these items paint a picture of rapidly maturing agentic capability alongside emerging adversarial threats to AI data integrity.
Implications for Australian agencies
- Monitor Agencies procuring AI systems trained on web-scraped data may want to monitor data-poisoning techniques like Poison Fountain as a supply-chain integrity risk.
- Consider Policy teams developing agentic AI governance frameworks could consider Drexler's institutional framing as a complement to existing system-level risk approaches.
Implications are AI-generated. Starting points, not advice.
"Import AI 441: My agents are working. Are yours?" Source: Import AI – Substack (Jack Clark) Published: 19 January 2026 URL: https://importai.substack.com/p/import-ai-441-my-agents-are-working This edition of Import AI covers three threads: a reflective essay by Anthropic's Jack Clark on his routine use of AI agents to complete research tasks autonomously; the emergence of 'Poison Fountain', an activist tool designed to corrupt AI training data by feeding poisoned content to web crawlers; and a summary of Eric Drexler's new framework arguing that AI governance should focus on institutions shaping many AI services rather than managing singular superintelligent agents. Together, these items paint a picture of rapidly maturing agentic capability alongside emerging adversarial threats to AI data integrity. Implications for Australian agencies: - [Monitor] Agencies procuring AI systems trained on web-scraped data may want to monitor data-poisoning techniques like Poison Fountain as a supply-chain integrity risk. - [Consider] Policy teams developing agentic AI governance frameworks could consider Drexler's institutional framing as a complement to existing system-level risk approaches. Retrieved from SIMS, 18 May 2026.