Thursday, May 14 on the Edge of AI. Apple's partnership with OpenAI is fraying, possibly into a legal fight. Claude just became the first model to clear all of Britain's cyberattack simulations. And ChatGPT's web traffic share has dropped twenty-four points in a year. Let's get into it.
Apple's two-year-old partnership with OpenAI, the lab behind ChatGPT, has reportedly become strained to the point that the AI startup is preparing possible legal action [1][11]. According to people familiar with the matter cited by Bloomberg, OpenAI hasn't seen the expected benefits from the deal and is now exploring litigation. The partnership, which powers intelligence features across Apple's ecosystem, was seen as a cornerstone of Apple's AI strategy. But the relationship has soured quietly. The signal: the largest consumer-electronics company and the most visible AI lab are now on a collision course over revenue-sharing and strategic alignment. That's a relationship that underpins a lot of the consumer-AI pipeline. If it breaks into a legal fight, the fallout touches every device that runs Apple Intelligence.
Different beat. Anthropic, the lab behind Claude, just shipped Claude Mythos Preview, and it's the first AI model to clear all cyberattack simulations from Britain's AI Safety Agency [2]. The UK's AI Security Institute, or AISI, had previously estimated that AI cyber capabilities were doubling every eight months. Then they revised that down to 4.7 months. Mythos has blown past even that accelerated timeline. Anthropic's head of red teaming, Logan Graham, warned, "Within a year, Mythos will probably look quite dumb." The model passed every single attack simulation AISI threw at it. Which matters because: the gap between frontier models and the rest is widening faster than regulators can measure. A model that clears the entire battery today is a model that, by the end of the year, may be able to autonomously execute novel cyberattacks. The safety community is watching this closely.
Now, on the web-traffic side. ChatGPT's share of generative-AI web traffic has dropped from 77.6 percent to 53.7 percent in twelve months [6][13]. Google Gemini is the biggest beneficiary, climbing from 7.3 percent to 26.7 percent. Claude went from 1.37 percent to 7.95 percent. The data, from Similarweb, only covers web traffic, not API usage or app downloads. But the trend is unmistakable: the market is fragmenting. ChatGPT still dominates, but its lead is shrinking by roughly two points a month. Gemini is absorbing most of that slack. The read: the consumer-AI market is no longer a single-player game. The window for new entrants is closing, but the window for established ecosystems like Google's is widening.
Three stories, one pattern: the AI landscape is getting more adversarial, more capable, and more fragmented all at once. The Apple-OpenAI legal risk, the Claude cyber milestone, and the traffic shift all point to a market that's maturing faster than its partnerships can keep up.
That is the edge for today.