Hopper agreed to pay $35M to the FTC on July 2, 2026 over alleged “dark patterns” that hid fees and misrepresented costs tied to “VIP Support” and “Price Freeze.” The FTC said “optional” charges were pre-selected and only surfaced after scrolling; Hopper must clearly disclose all fees before users finish a transaction.
The Department of Homeland Security is investigating a breach of its Homeland Security Information Network (HSIN), a platform used by federal, state, and local governments and law enforcement to…
A group of AI researchers launched FLARE-AI (Flaw Reporting for AI) in 2026, a crowdsourced website designed to let users report and track harmful behavior from AI systems — from chatbots generating…
A security researcher claims to have discovered a bug in Apple’s Hide My Email feature that allows users’ real email addresses to be unmasked, potentially undermining the privacy tool entirely.
The…
Cloudflare says AI firms have until Sept. 15, 2026 to split “mixed-use” crawlers for traditional search from those used for AI training/agents, or face automatic blocks on publisher sites. From that date, default settings block them on ad pages for new customers, new sites, and all free-tier sites; owners can change settings.
On July 1, 2026, Venice AI says its client-side encryption and external proxy mean user input isn’t stored on its systems, even as it runs 200+ open-source models and routes to OpenAI/Anthropic. The $65M Series A led by Dragonfly values it at $1B; it’s profitable with $70M+ run-rate revenue, 3M+ active users, and 1.7M API calls/day.
A security researcher discovered in April 2026 that Anthropic’s AI tool Claude Opus 4.7 could help him break into Front Gate Tickets’ systems and freely issue tickets to nearly every major US music…
Anthropic launched Claude Sonnet 5 on June 30, 2026, positioning the mid-tier model as a lower-cost alternative to its own Opus 4.8 as well as OpenAI’s GPT-5.5 and Gemini 3.1 Pro. The model is now…
Proton launched Lumo 2.0 on June 30, 2026, bringing image recognition/generation, “thinking mode,” and up to 76% faster replies under zero-access encryption. Projects can upload docs across Proton email and cloud storage with user-controlled persistent memory; Proton also says it keeps no server-side session logs and never uses data for AI training or shares it.
On June 30, 2026, X launched a hosted MCP server so AI apps can call X’s API using a user’s own account permissions, avoiding self-hosting and auth work. Tools like Claude, Cursor and Grok Build can reuse existing search, post-reading, user lookups, and conversation analytics, with spam controlled by X’s API rules. X joins GitHub, Slack, Notion, Stripe, and Salesforce.
WIRED says Meta’s Covalen-managed “Cannes” project had hundreds of contractors impersonate minors and probe ChatGPT, Gemini, and Character.AI with prompts on suicide, self-harm, eating disorders, sex/romance, and drugs. It ran into Apr. 21, 2026; one Aug. 2025 round topped 45,000 prompts, and WIRED reviewed 3,748 prompts including at least 239 covering sex/romance.
California Governor Gavin Newsom and Anthropic announced in June 2026 a deal giving state and local government agencies access to Claude, Anthropic’s AI chatbot, at half price. The agreement also…
