AI news · Saturday, October 17, 2026
Websites are aggressively blocking AI agents to protect their own traffic
Websites from Delta Air Lines to JPMorgan Chase and even Amazon are increasingly blocking autonomous AI agents from accessing their platforms. This isn't just about security concerns; it's a defensive move to keep human eyeballs on their sites, where companies can actually show ads and sell things. Meta’s agent, Muse, has already faced takedowns from Amazon, forcing Meta to pivot toward formal partnerships with retailers like Walmart and Best Buy to ensure its agents can actually shop.
Meanwhile, in the corporate world, Anthropic is finding that its own reliance on Claude is fundamentally changing how it operates. The company reported that its workforce has shifted away from junior talent, as coding tools allow non-technical staff to build software themselves. This has left the remaining team spending 20% more time in meetings compared to last year, as human judgment becomes the bottleneck in a workflow saturated by AI-generated output.
In the coding space specifically, Replit is pulling back from its Anthropic partnership in favor of OpenAI. The platform’s CEO, Amjad Masad, cited performance and cost as the primary drivers, claiming ChatGPT currently offers better results for building applications from scratch. Despite this, Replit maintains it is still model-agnostic, though the shift to OpenAI as the default for its AI agent will likely dent Anthropic’s revenue significantly.
Meta is also walking back its own aggressive data grab after a PR headache. The company removed language from its terms of service that seemed to grant it broad permission to scrape any publisher's website connected to a Meta business account. Meta insists this was just a clarification and not a change in how its crawlers operate, but the retreat follows a pattern of companies trying to balance AI training needs with mounting legal and public pressure.
Finally, the promise of AI solving the deepest mysteries of biology is hitting a reality check. While foundation models like Google DeepMind’s AlphaGenome are useful for predictions, experts warn that the genome is not a static blueprint—it’s a dynamic, context-dependent system that current AI models aren't equipped to fully understand. Without the data to map out how genes respond to environmental signals in real time, these models remain computational black boxes that may never truly grasp the complexity of life.
The quick hits
- Websites like Delta, JPMorgan, and Amazon are blocking AI agents to protect their traffic and security — it's a major roadblock for the 'personal assistant' future AI companies promised.
- Anthropic says its employees are spending 20% more time in meetings than last year — turns out, AI creates so much content that humans are becoming a bottleneck for review.
- Replit is dumping Anthropic as its primary partner in favor of OpenAI — the move is a blow to Anthropic's developer-focused strategy and signals a fight for market dominance.
- Meta backtracked on terms that allowed it to scrape business websites for AI training — the company claims it was a misunderstanding, but publishers are watching their data carefully.