AI This Week: Patreon Blocks Training Bots, GPT-Red Attacks Models, and Google Puts AI Into Biosecurity

Patreon says its Cloudflare crawler block cut thousands of weekly training-bot attempts to zero, while OpenAI’s GPT-Red beat human red-teamers 84% to 13%.
artificial-intelligence
cyber-security
Author

Kabui, Charles

Published

2026-07-19

Keywords

patreon-ai-crawlers, gpt-red, prompt-injection, google-bioresilience, grok-automations

Download as PDF

Patreon began blocking known AI training crawlers at the network level with Cloudflare, while still allowing search crawlers that send readers back to creators. Patreon says early tests cut individual training bots from thousands of weekly access attempts to zero, but no independent audit backs that figure. OpenAI also introduced GPT-Red, an internal model trained through self-play to attack other models with prompt injections. In OpenAI’s held-out test, GPT-Red found successful attacks in 84% of scenarios, compared with 13% for human red-teamers. GPT-5.6 Sol now fails on 0.05% of GPT-Red’s direct attacks, according to OpenAI.

Google DeepMind and Isomorphic Labs reported more than 15 biosecurity partnerships in 12 months. Their plans include adapting SynthID to mark AI-generated biological sequences and using AlphaEvolve to improve outbreak detection. Google Vids added prompt-based video editing and personal avatars for paid AI and Workspace users, with age and regional limits. SpaceXAI released Grok Automations for scheduled jobs, while email triggers require SuperGrok. It also published Grok Build’s coding-agent harness under Apache 2.0, including its agent loop, tools, terminal interface, and local-inference support.

Put together, the releases show why background AI needs stronger controls. A scheduled agent can read email and use connected tools without a live prompt. A biology model can help with outbreak response, but generated sequences also need screening. Crawler blocks, attack testing, access limits, and provenance marks are becoming part of the product.

Read More: AI This Week: Grok 4.3 Cuts Prices, Claude Security Launches, Pentagon Signs Eight AI Deals

Sources:

Disclaimer: For information only. Accuracy or completeness not guaranteed. Illegal use prohibited. Not professional advice or solicitation. Read more: /terms-of-service

Reuse

GNU GENERAL PUBLIC LICENSE v3.0(View License)

Citation

BibTeX citation:
@misc{kabui2026,
  author = {{Kabui, Charles}},
  title = {AI {This} {Week:} {Patreon} {Blocks} {Training} {Bots,}
    {GPT-Red} {Attacks} {Models,} and {Google} {Puts} {AI} {Into}
    {Biosecurity}},
  date = {2026-07-19},
  url = {https://toknow.ai/posts/ai-weekly-patreon-gpt-red-biosecurity-grok-automations/},
  langid = {en-GB}
}
For attribution, please cite this work as:
Kabui, Charles. 2026. “AI This Week: Patreon Blocks Training Bots, GPT-Red Attacks Models, and Google Puts AI Into Biosecurity.” https://toknow.ai/posts/ai-weekly-patreon-gpt-red-biosecurity-grok-automations/.