Anthropic's Mythos is too dangerous for the open market but too useful to be ignored. Though only released to a relatively small number of large-scale developers in order for them to develop defenses to Mythos, the hyper-powerful agentic-organizer appears to have itself been hacked. A blog post from the Mozilla Foundation noted Mythos helped their developers find and close 277 security vulnerabilities in a new product release, a figure far higher than the average of 25 flaws per release. The Trump regime appears to be softening its stance against Anthropic with the President telling CNBC a deal allowing the Department of Defense to continue use of Anthropic Claude's intelligence models.
The Internet Archive, better known as the WayBackMachine, is being threatened as large news gathering organizations are disallowing Internet Archive crawlers from accessing their sites. The problem is AI crawlers are using the Archive as a way to access news content they are otherwise not able to crawl. The Internet Archive is the largest independent archive of content that's appeared on the Internet. It's preservation is critically important.
Jim and Kristine talk about the evolutions of SEO education from the early days to today in response to statements about SEO from Google's John Mueller, one of which suggested people who self-reference with the word "Guru" are likely "clueless imposters". They also talk about Google VP of Search Liz Reid's views on how AI Overviews are changing search, how a Roblox cheat download opened the door for the hack at Vercel, how the power/energy/token crisis could limit what lower-tier AI products offer, a number of new Google and MSFT Ads features, and a lot of information from Google's Search Central Live event in Toronto.
It's a longer, fun, and a totally non-commodity set of conversations on this week's Webcology.