> _Your fast-read digest of the most important developments in AI this week._ ##### TL;DR: Give me the [[#💼 Impact & Opportunities for Salesforce Consultants]] --- > **Editor's Note:** There may not have been a single week in the history of AI quite like this one. Google I/O reshaped the subscription and agent landscape. OpenAI filed its IPO. Anthropic reported its first-ever profit and hired the most beloved AI educator alive. A hacker group hit GitHub, OpenAI, and Mistral through a poisoned VS Code extension. An AI autonomously solved an 80-year math problem. The Pope weighed in. And the White House cancelled its own AI safety order after three tech billionaires called the president directly. Eight stories follow — each one would have led the news on a quiet week. --- ## 1. Google I/O 2026 — The Most AI-Packed Keynote in Google's History Google CEO Sundar Pichai opened the I/O 2026 keynote on May 19 with scale numbers the industry had never heard before: **3.2 quadrillion tokens processed monthly**, **900 million Gemini app monthly active users**, **1 billion AI Mode monthly active users**. DeepMind CEO Demis Hassabis then took the stage and stated flatly: "Artificial General Intelligence is just a few years away." What followed across the two-hour keynote was the broadest single-day product launch in Google's history. The headline announcements, each significant on its own: - **Gemini 3.5 Flash** — available immediately, now powering Google Search AI Mode, Antigravity 2.0, and all AI Overviews. Gemini 3.5 Pro is in testing and arrives next month. - **Gemini Omni** — a new unified text, image, audio, and video model available today for paid Gemini subscribers. It combines Gemini's reasoning with Nano Banana (image generation) and Veo (video) in a single pipeline. Conversational video editing via text prompts, with higher fidelity and embedded background music generation. - **Gemini Spark** — Google's 24/7 personal AI agent running on Google Cloud VMs even when your laptop is closed. It works autonomously across Gmail, Sheets, Slides, Drive, and Calendar, surfacing progress via Android Halo. Launches next week for US AI Ultra subscribers, with MCP support for third-party apps coming in weeks. - **Google AI Ultra drops from $250 to $100/month** — the most consequential AI subscription pricing move of 2026. The $100 plan includes 5x higher usage limits, 20TB storage, YouTube Premium, and Gemini Spark access. Daily prompt limits are replaced by a compute-based model refreshing every five hours. - **Ask YouTube** — conversational AI search inside YouTube, live today for Premium users, turning the world's largest video library into a queryable knowledge base. - **Universal Cart** — AI-powered shopping across Search, Gemini, YouTube, and Gmail, integrated with Amazon, Shopify, and Walmart via the new Universal Commerce Protocol. - **Samsung Android XR Smart Glasses** confirmed for fall 2026, with Gentle Monster and Warby Parker frames, two tiers (audio-only and in-lens display), priced between $379–$499, compatible with both Android and iPhone. 📖 [Read More — Build Fast with AI: May 20 AI News](https://www.buildfastwithai.com/blogs/ai-news-today-may-20-2026) | [Read More — Unrot: Weekly AI Update May 19–24](https://unrot.co/blogs/weekly-ai-news-may-2026) --- ## 2. Andrej Karpathy Joins Anthropic — The Biggest Talent Move of 2026 On May 19 — the same day as Google I/O — Andrej Karpathy announced he is **joining Anthropic**. He starts immediately on the pretraining team under Nick Joseph, with a mandate to build a new team focused on AI education and making Claude more accessible. Karpathy, creator of the massively popular Neural Networks: Zero to Hero course on YouTube and a co-founder of OpenAI, is widely regarded as the most effective AI educator in the world. His arrival at Anthropic signals two things simultaneously: Anthropic is investing heavily in the developer and education ecosystem around Claude, and the talent market continues to consolidate around the labs with the strongest technical reputations. 📖 [Read More — Unrot: Weekly AI Update May 19–24](https://unrot.co/blogs/weekly-ai-news-may-2026) --- ## 3. OpenAI Files for IPO — And Wins the Musk Lawsuit OpenAI filed its S-1 registration statement with the SEC on Friday, May 23, 2026, formally beginning the IPO process. The three-company AI IPO wave now has a confirmed timeline: **SpaceX in June, OpenAI in September, Anthropic in October** — the most concentrated cluster of near-trillion-dollar technology listings in capital markets history. Bloomberg and the Financial Times simultaneously confirmed that Anthropic is on track to close its own **$30 billion round at a $900 billion valuation** as soon as the week of May 26, co-led by Sequoia, Dragoneer, Altimeter, and Greenoaks. The week also delivered a major legal clearing for OpenAI: a California jury needed less than two hours to **unanimously reject Elon Musk's entire lawsuit** against the company. Musk had claimed OpenAI violated its original nonprofit mission; the jury found no merit in the argument. The verdict removes the largest outstanding legal cloud over OpenAI's IPO. Separately, Anthropic disclosed its **first quarterly operating profit** — a milestone that validates the revenue trajectory behind the $900B valuation and provides the kind of P&L evidence that institutional IPO investors require. 📖 [Read More — Unrot: Weekly AI Update May 19–24](https://unrot.co/blogs/weekly-ai-news-may-2026) | [Read More — Build Fast with AI: May 23 AI News](https://www.buildfastwithai.com/blogs/ai-news-today-may-23-2026) --- ## 4. OpenAI's AI Autonomously Solves 80-Year-Old Math Problem — Fields Medalists Take Notice OpenAI announced on May 20 that one of its internal general-purpose reasoning models **autonomously disproved the Erdős Unit Distance Conjecture** — a central problem in discrete geometry that Paul Erdős posed in 1946. The result is a 125-page proof establishing an infinite family of planar point configurations that produce significantly more unit-distance pairs than the square-grid arrangements mathematicians had long considered optimal, refuting Erdős's conjectured upper bound entirely. What surprised the mathematics community was not just the result but the method: the model approached the problem through **algebraic number theory**, connecting an elementary geometry question to deep number-theoretic structures that combinatorial geometers had never thought to apply. It was not guided step-by-step by humans and did not retrieve an existing solution. **Fields medalist Tim Gowers** reviewed the work and called it "a milestone in AI mathematics." Princeton mathematician Noga Alon described it as "an outstanding achievement" that "applies fairly sophisticated tools from algebraic number theory in an elegant and clever way." This is the first time AI has autonomously solved a prominent open problem central to a field of mathematics — not a competition benchmark, but real frontier research. 📖 [Read More — Build Fast with AI: May 23 AI News](https://www.buildfastwithai.com/blogs/ai-news-today-may-23-2026) --- ## 5. Anthropic Co-Founder Predicts Nobel Discovery in 12 Months — and Extinction Risk Isn't Gone Anthropic co-founder **Jack Clark** delivered the 2026 Cosmos Lecture at Oxford on May 20, titled "Change is inevitable. Autonomy is not." His predictions have been circulating ever since. Clark stated he believes AI will work with humans to make a **Nobel Prize-winning discovery within 12 months**. He forecast bipedal robots assisting tradespeople within two years. He predicted companies **run entirely by AI will be generating millions in revenue within 18 months**. The most discussed prediction: a **60%+ chance that an AI model will fully train its successor by end of 2028** — recursive self-improvement, what Clark called an "intelligence explosion" — now in official Anthropic research documents, not just theoretical speculation. He explicitly acknowledged a "non-zero chance" that AI could kill everyone on the planet, compared the lack of preparation to the failure to prepare for COVID-19, and called for pandemic-style institutional readiness. Clark was speaking as Anthropic simultaneously approaches a $900B valuation and reports its first quarterly profit — these are not abstract philosophical views but Anthropic's actual internal assessment of what the next 24 months will bring. 📖 [Read More — Build Fast with AI: May 23 AI News](https://www.buildfastwithai.com/blogs/ai-news-today-may-23-2026) --- ## 6. Meta Caught Training AI on Employees Before Firing 8,000 of Them A **leaked audio recording from a Meta all-hands meeting** on April 30 surfaced publicly on May 19 — the exact day approximately 8,000 Meta employees received layoff notices. In the audio, CEO Mark Zuckerberg defends what Meta calls the **"Model Capability Initiative"**: an internal program that tracks employee activity across Gmail, Google Chat, the internal assistant Metamate, and VS Code to train Meta's AI models on how smart people work. "The AI models learn from watching really smart people do things," Zuckerberg is heard saying. The audio dropped at the moment those 8,000 employees were being notified of their terminations — rolling through Singapore at 4 AM, then Europe, then the US. Employees organized internal protests, pasting flyers in meeting rooms urging colleagues to sign petitions against the initiative. The phrase "Black Mirror episode" appeared in hundreds of social posts. Meta has committed more than **$125 billion to AI infrastructure and data centers in 2026** — Zuckerberg's framing is that smaller, AI-assisted teams will do more. Workers' framing is that they helped build the system that eliminated their jobs. 📖 [Read More — Build Fast with AI: May 23 AI News](https://www.buildfastwithai.com/blogs/ai-news-today-may-23-2026) --- ## 7. Trump Kills AI Safety Executive Order — After Direct Calls from Musk, Zuckerberg, and Sacks The White House AI executive order — a voluntary 90-day pre-launch review framework for frontier AI models with NSA involvement in classified testing — was **scrapped on May 21, hours before the scheduled signing ceremony**. Invitations had already been sent. Axios obtained the explanation: the main reason was that Trump "just hates regulation," and former AI czar David Sacks "hated it" too. Between Wednesday night and Thursday morning, **Sacks, Elon Musk, and Mark Zuckerberg all spoke with Trump directly** — outside the normal policy process — warning the framework would slow AI development. Trump publicly explained: "I didn't like certain aspects of it. I think it gets in the way of — we're leading China, we're leading everybody, and I didn't want to do anything to get in the way of that lead." The real story: **informal CEO access to the president outweighed months of interagency work** on a critical security question. The order was originally prompted by Anthropic's Claude Mythos discovering zero-day vulnerabilities at scale. The national security professionals who spent weeks building the draft had no comparable access to the decision. 📖 [Read More — Build Fast with AI: May 23 AI News](https://www.buildfastwithai.com/blogs/ai-news-today-may-23-2026) --- ## 8. GitHub, OpenAI, and Mistral Hacked via Poisoned VS Code Extension — 3,800 Repos Stolen GitHub confirmed on May 20 that approximately **3,800 internal repositories were exfiltrated** by threat actor group **TeamPCP** (also tracked as UNC6780). The attack vector: a trojanized version of the **Nx Console VS Code extension** — 2.2 million installs, verified publisher status — that was live on the Visual Studio Marketplace for exactly **18 minutes** between 12:30 and 12:48 PM UTC on May 18. The malicious version executed silently on startup, downloading a credential stealer that targeted 1Password vaults, Anthropic Claude Code configurations, npm tokens, GitHub tokens, and AWS credentials. The GitHub breach was one stop in a wider attack: **TeamPCP also hit OpenAI, Mistral, the European Commission, and 170+ npm packages**. The operational security implication is stark: **the fastest confirmed developer supply chain attack ran 18 minutes from deployment to credential harvest**. Standard endpoint detection is not designed for that timeline. VS Code extensions, npm packages, and PyPI distributions operate on a trust layer that traditional perimeter security does not cover. 📖 [Read More — Build Fast with AI: May 23 AI News](https://www.buildfastwithai.com/blogs/ai-news-today-may-23-2026) --- ## 💼 Impact & Opportunities for Salesforce Consultants The week ending May 23 was not just the biggest AI news week of 2026 — it was arguably the biggest in the industry's history. For Salesforce consultants, the signal that matters most is not any single story but the pattern across all of them: **AI is simultaneously becoming the infrastructure of enterprise workflows, the target of enterprise security threats, and the driver of the most consequential pricing and talent shifts in the technology market.** Here is what that means in practice. ### ⚠️ Role Impact: How Your Work Is Changing - **Google AI Ultra at $100/month with a 24/7 agent changes client expectations overnight.** When the world's most integrated AI suite drops to $100/month and includes a background agent that works across Gmail, Sheets, Drive, Calendar, and Slides while you sleep, clients will immediately benchmark their Salesforce experience against it. Every Agentforce implementation that requires manual configuration, human-triggered workflows, or doesn't surface proactive intelligence will feel dated relative to Gemini Spark's demo. The bar for what "agentic" means just moved — visibly and publicly. - **The supply chain attack on developer tooling hits Salesforce implementations directly.** The TeamPCP attack targeted Claude Code configurations, npm tokens, and GitHub credentials — exactly the toolchain most advanced Salesforce development shops use for CI/CD, metadata deployment, and Apex development. If your team uses VS Code with Salesforce CLI extensions, npm packages for LWC development, or GitHub for Salesforce DX pipelines, the attack vector is real and active. Consultants who cannot speak to software supply chain security are not ready for the enterprise security conversations clients will be having next week. - **Andrej Karpathy's move to Anthropic is a talent signal for the whole ecosystem.** When the most effective AI educator in the world leaves to build the education layer around Claude, it signals that Anthropic's next phase is about developer fluency and accessibility — not just raw capability. This will accelerate Claude adoption among the developer population that builds on Salesforce. Consultants who are not building Claude fluency now will be competing against developers who are being taught by the best teacher in the field. - **OpenAI's math breakthrough reframes what AI agents can do unsupervised.** An 80-year research problem solved without human guidance is not just a benchmark — it is a data point about autonomous AI judgment at a level that exceeds expert human capability in a specific domain. For Salesforce consultants, the implication is not abstract: the same reasoning architecture that solved the Erdős problem is the architecture running inside Agentforce agents making autonomous decisions in your clients' orgs. The governance, audit, and explainability conversations just became more urgent. ### ✅ Opportunities: Where to Play Offense - **Reposition your Agentforce implementations around the Gemini Spark comparison.** Clients who see Gemini Spark's demo — a 24/7 agent working autonomously across every Google Workspace tool — will ask why their Salesforce experience doesn't feel the same. Use that comparison productively: help clients map which of their workflows can be handled by Agentforce today (CRM-centric, data-rich, structured processes) and which are better served by Gemini Spark (cross-app, document-heavy, personal productivity). Consultants who can give a clear, honest matrix win the strategic advisor role. - **Build a developer supply chain security assessment offering now.** The TeamPCP attack is exactly the kind of event that prompts CISOs to demand an audit of every tool in the development pipeline. For Salesforce-focused development shops, that means reviewing VS Code extension permissions, auditing npm dependencies in LWC and Salesforce DX projects, reviewing GitHub Actions configurations, and implementing signed commit policies. This is a concrete, billable engagement that can be delivered immediately — and one that most Salesforce implementation teams have never done. - **Use Google AI Ultra's pricing as a lever in your own tooling strategy.** A $100/month subscription with frontier video generation, a 24/7 agent, and 20TB of storage is the most cost-efficient AI tooling on the market. Build internal workflows that combine Google Workspace (now fully Gemini-native) with your Claude and Salesforce tooling — client research, meeting prep, documentation drafting, proposal generation. Consultants operating with an integrated AI toolkit across Gemini, Claude, and Agentforce will consistently out-produce single-tool users. - **Lead governance conversations that Jack Clark's Oxford predictions make unavoidable.** When an Anthropic co-founder publicly states there is a 60%+ chance of recursive AI self-improvement by 2028 and a non-zero chance of extinction-level risk — and does so while his company approaches a trillion-dollar valuation — enterprise risk committees and board audit functions will start asking their technology advisors what their AI governance framework looks like. Salesforce consultants who show up with a structured point of view on Agentforce governance, audit trails, and human-in-the-loop design are solving a problem that just moved from the "future concern" bucket to the "boardroom agenda" bucket. - **Use the IPO wave as a business development trigger.** Three near-trillion-dollar AI IPOs in a single quarter — SpaceX, OpenAI, Anthropic — will generate more C-suite AI strategy conversation in the next 90 days than any product launch could. Every major enterprise client will be reading about the AI valuation wave and asking what their own AI strategy is worth. Consultants who show up in that conversation with a clear point of view on what Agentforce + Claude integration means for their specific client's industry will win the engagements that follow. --- _Brief compiled May 23, 2026 | Sources: Build Fast with AI, Unrot, Digital Applied_