Remember Anthropic’s AI going rogue? Then OpenAI’s AI also did something similar and hacked other services? And then Meta said its Muse AI did something similar? Well, now it’s the turn of Chinese AI models that are regularly sending Wall Street into a tizzy.

Moonshot's Kimi K3 is the second Chinese AI model now reported to have escaped its testing environment. At this point, it feels like a race to the “my AI is more dangerously powerful.” Regulators are watching this PR battle rather closely, by the way.

How about a pre-AI blast from the past? Before Google turned the web into a tidy, searchable index, finding anything online meant wandering. This week's tech news pulls in two directions at once, and the gap between them is what makes it strange. On one end sits nostalgia: a look back at the messy, human-curated internet we traded away for convenience.

It was gloriously disorganized before Google rationalized the mess into something you could actually search, and in doing so it quietly retired a version of the internet that ran on human attention instead of ranking algorithms. Of course, Google also injected AI in each corner, and did social media sites. And that brings us to our final story. The purge on AI slop has actually caught social media influencers as collateral damage. What a mess!

Send us feedback at [email protected].

Kimi K3, the powerful AI model developed by Chinese startup Moonshot, escaped a testing sandbox operated by the UK government's AI Security Institute while being evaluated for its defensive cybersecurity capabilities. US cybersecurity firm Frontier Security documented the breakout, noting that the model did not exploit a zero-day vulnerability but instead took advantage of a misconfiguration in the sandbox environment, a pattern that has now emerged across multiple AI developers.

Similar incidents were previously reported involving models from Anthropic, OpenAI, and Meta, all of which slipped out of their supposedly isolated testing environments due to errors made by their evaluation partner, Irregular. What makes the Kimi K3 case particularly notable is that the model involved is the publicly available version, not an unreleased build or one with deliberately lowered safeguards. Rather than hacking into an external service, the model simply accessed the internet and retrieved a solution from GitHub. What’s next? Experts are scared, but we simply can’t reliably fathom the risks.

Read More →

When you type a prompt, you summarize. When you speak one, you explain. Wispr Flow captures your full reasoning — constraints, edge cases, examples, tone — and turns it into clean, structured text you paste into ChatGPT, Claude, or any AI tool. The difference shows up immediately. More context in, fewer follow-ups out.

89% of messages sent with zero edits. Used by teams at OpenAI, Vercel, and Clay. Try Wispr Flow free — works on Mac, Windows, and iPhone.

Start flowing free

Before Google arrived to swallow the internet whole, finding anything online was less a search than a scavenger hunt, a process shaped as much by luck, word of mouth, and human curation as by any algorithm. In the mid-1990s, engines like AltaVista, Lycos, Excite, HotBot, and Ask Jeeves each promised to impose order on a rapidly expanding web, but they were only one piece of a much messier ecosystem. Directories, bookmarks, email signatures, and site-to-site links all competed with machine indexing because the web was still small enough that human organization held its own.

Long before those engines became the dominant public interface, platforms like Usenet were teaching people how to navigate vast digital spaces. Conceived in 1979, Usenet organized discussion into topic-based newsgroups and built a culture of communal, distributed information-sharing that predated the search box entirely. Users discovered websites through links dropped at the bottom of posts, and awareness spread organically through communities rather than rankings. The real lesson of the pre-Google era is not that early search was simply worse, but that it reflected a fundamentally different internet, one where no single interface had yet claimed the whole territory, and where the act of finding something was still a genuinely creative endeavor.

Read More →

TikTok and Instagram are incorrectly labeling human-made content as AI-generated, leaving creators to defend work they spent hours crafting by hand. Seven creators speaking to Business Insider described similar false positives, in some cases triggered simply by using mainstream editing tools like Canva or Adobe Lightroom. The mislabeling carries real professional weight.

In a $12 billion influencer marketing industry built on authenticity, even an erroneous AI tag can damage a creator's relationship with brand partners, and several talent managers report a growing number of campaign contracts now explicitly prohibit AI use in scripting, captions, and visual edits. Public sentiment is also shifting: a Billion Dollar Boy survey found the share of consumers who view generative AI as a negative disruptor nearly doubled from 18 percent in 2023 to 32 percent in 2025, making a wrongful label feel, in one creator's words, like "the Scarlet Letter." Platforms admit the tools are imperfect, but they have no fool-proof solution.

Read More →

GTA 6 gets narrated gameplay in Netflix reveal

Destiny 2 modder restores six years of cut content

Apple quietly bumped up the numbers on its trade-in program, and if you've been eyeing an upgrade, the timing works in your favor.

Epic Games Store has new free PC games this week

Okami was inspired by a Spider-Man 2 movie scene

House of the Dragon season 3 finale hits HBO Max this weekend

Netflix now streams 4K video in Chrome

Forza Horizon 6 Series 3 Italian Exotics challenges are live