Strategic Implications for Business Leaders

Welcome to Future Bytes News, your weekly briefing on AI and business. I want to start this week with two events that happen three days apart. When you put them next to each other, you get the most consequential week we've covered on this show this year. On Monday, May 25th, Pope Leo 14th released his first encyclical, the highest form of papal teaching. It runs 42,000 words. It's called Magnifica Humanitas, Magnificent Humanity, and the entire document is about artificial intelligence. The Pope's position is direct. Quote Artificial intelligence now demands to be disarmed, freed from logics that turn it into an instrument of domination, exclusion, and death. He wrote that technology is never neutral because it takes on the characteristics of the people who design, finance, regulate, and use it. He warned specifically against the concentration of data and infrastructure in a handful of corporations. He said the pursuit of greater profits cannot justify choices that systematically sacrifice jobs. And he called for robust legal frameworks and independent oversight to make sure AI serves the common good. Now here's the detail that makes this week land. The Pope presented the encyclical at the Vatican that Monday, and standing alongside him as a symbolic gesture of dialogue between the Church and the AI industry was Christopher Ola. Ola is a co-founder of Anthropic. Three days later, on Thursday, May 28th, Ola's company made three announcements. The first one is the one to slow down on because it 's a reversal. Seven weeks ago on April 7th, Anthropic told the world it had built an AI model so capable of finding and exploiting software vulnerabilities autonomously at scale across every major operating system and web browser that they were not going to release it. They called it mythos . The UK government's AI security institute called the capability on charted territory. The US Treasury convened em ergency meetings about it. Anthrop ic restricted the model to about 50 defensive security organizations under a program called Project Glasswing. It was the first time a major AI lab had built a frontier capability and decided not to sell it. On Thursday this week, Anthropic announced their releasing Mythos class capabilities to all customers in the coming weeks. The reason they gave is what they called a seven-week safety sprint focused on making models more honest about their own limitations. And it's worth quoting their own caveat directly. Anthropic itself said, At present, no company including Anthropic has developed safeguards strong enough to prevent such models from being misused and potentially causing severe harm. That's the company telling you the safeguards are not yet strong enough, and telling you the model is coming to everyone in a matter of weeks. The second Thursday announcement was the funding round to pay for it. Anthropic close to sixty-five billion dollars Series H at a nine hundred and sixty-five billion dollar valuation. That makes them the most valuable AI startup in the world. I'm not telling you that as investor news, I'm telling you because it means whichever vendor your roadmap depends on, both of the leading labs now sit on enough capital to outspend most national AI programs combined. That changes how you think about vendor durability over a three to five year planning horizon. And the third announcement on Thursday was a new flagship model called Claude Opus 4.8. Better at agentic coding, better at legal reasoning, and four times less likely than the previous version to let a flaw in its own code pass unflagged. It's the everyday model, and it's available wherever your engineers already use, Claude. So pull all of that back for a moment. On Monday, the Pope stood at the Vatican next to an anthropic co-founder and said AI must be disarmed and warned against the concentration of AI power in a handful of corporations. On Thursday , that co-founder's company announced the wide release of the most powerful cybersecurity AI ever built, the largest private AI funding round in history, and a new everyday flagship model. Both events were real, both happened in the same five business days. Both belong on your roadmap. Now here's why all of this matters for your business. The mythos reversal is the part that I think most security teams have not yet absorbed. The exact same capability that was deemed too dangerous to release seven weeks ago is about to be available in the same API your engineers are already calling. That changes your threat model in a specific way. Mythos class systems find vulnerabilities. They don't care which side of a firewall they're on. So when the model ships, the defenders get it. And so does anyone else who can buy an API key. That symmetry, defenders and attackers receiving the same capability on the same day is what your security team should be planning around, not whether the safeguards anthropic just described are strong enough. Anthropic told you they're not. The practical action is straightforward and I do it this month. Get your CISO, your head of engineering and your AI procurement lead in the same room before the model ships. Map two things. First, what does your patch cycle look like when a missile Second, what does your incident response plan look like the day after a competitor or an adversary calls the same API. Now let me widen the frame because Anthropic was not the only lab making access decisions this week. They were moving in one direction. OpenAI was moving in the other. On Friday, May 29th, OpenAI announced expand ed but still tightly gated access to a model called GP to Rosalind for biodefense, public health, and pandemic preparedness work. The recipients are vetted developers and US government partners. Same day, OpenAI published a frontier governance framework. So this week, the two leading labs in the world moved in opposite directions on the same fundamental question. Should a frontier capability be widely released or tightly gated. They are not converging on a shared safety standard. Which means your vendor diligence has to assess each one separately and keep assessing. And there's one more thread worth pulling. Open AI's confidential IPO filing was reported publicly this week, with a listing targeted for the fourth quarter, possibly as early as September. So within six months, one of the two leading AI vendors becomes a public company with audited financials. The other stays private. That asymmetry matters for how you do vendor diligence. The inputs you have on each of them are about to be different in kind . A few more developments worth noting briefly. Meta on Wednesday launched paid consumer subscriptions across its social platforms and signaled a dedicated AI subscription tier as coming. That's the first sign of a major social platform funding its AI build by charging consumers directly rather than relying entirely on ads. If your business advertises on those platforms, the audience monetization model your media plan rests on is starting to shift. XAI also added another coding model to the increasingly crowded developer tooling stack. Worth a quick look if you're benchmarking developer tools. And Microsoft Build runs Tuesday and Wednesday next week, so next week's episode will cover that keynote. Expect a heavy focus on AI agents, Microsoft's own coding model, and the next phase of Copilot. For your AI roadmap , four practical questions to bring to your next planning session. First, if Mythos class vulnerability finding capability is going to be commercially available in a matter of weeks, when does your security team's threat model get updated ? And by whom? Get the meeting on the calendar before the model ships. Second, is your AI vendor diligence keeping pace with the divergence between the labs? This week one of them loosened access on a dangerous capability and the other tightened access on a different one. The labs are not on a shared standard. You have to assess each one separately and you have to keep assessing. Third, does your roadmap account for one of your vendors becoming a publicly traded company in the next six months while the other stays private with the deepest war chest in the industry? The disclosure profile of open AI changes the moment they list. Anthropics stays opaque. Your inputs for evaluating each one are about to be different in kind. Fourth, have you mapped where mythos class capabilities would show up in your business if the defenders had them? And if attackers had them on the same day. That symmetry is the real practical question, because the same wide release reaches both. The theme across all of this is clear. Seven weeks ago, an AI lab told the world it had built something so capable of finding software flaws at scale that releasing it would be dangerous. This week the same company told the world it's coming to everyone. And they raised the largest private funding round in industry history to make it happen. And on Monday, three days before that announcement, the Pope stood at the Vatican and said the technology must be disarmed and warned specifically against the concentration of AI power in a handful of corporations. Both things were true in the same week. Both belong on your roadmap. For business leaders, the honest read is direct. Your assumption that frontier AI access decisions stays stable for more than a quarter has just been falsified in writing. The capability that triggered emergency treasury meetings in April is about to be commercially available in the same APIs your engineers already use. Your security team should know that before the model ships. Your AI vendor diligence should account for two labs moving in opposite directions on access. And the moral framing, the kind of framing the Pope put on it