> _Your fast-read digest of the most important developments in AI this week._ ##### TL;DR: Give me the [[#💼 Impact & Opportunities for Salesforce Consultants]] --- ## 1. Anthropic's $30B Funding Round Officially Closes — World's Most Valuable Private AI Company The deal that has been building for weeks became official: Anthropic's **$30 billion-plus funding round closed during the week of May 26**, confirming it as the world's most valuable private AI startup — surpassing OpenAI's $852 billion March 2026 private valuation for the first time. The round was co-led by **Sequoia Capital, Dragoneer Investment Group, Altimeter Capital, and Greenoaks Capital Partners** (roughly $2B each), with **Microsoft, NVIDIA, Founders Fund, and General Catalyst** also participating. It came together in under four weeks — unusually fast by Bloomberg's account. The revenue trajectory behind the valuation tells the real story: Anthropic grew from $87M ARR in January 2024 to $1B by December 2024, $9B by end-2025, and **$30B in April 2026** — driven overwhelmingly by Claude Code and enterprise adoption. The round is expected to be Anthropic's final private capital injection before an October 2026 IPO targeting. Separately, Anthropic has arranged a **$45B compute deal with SpaceX** and a **$1.8B agreement with Akamai Technologies** for additional infrastructure. 📖 [Read More — Build Fast with AI: May 27 AI News](https://www.buildfastwithai.com/blogs/ai-news-today-may-27-2026) | [Read More — AI-Weekly Issue 218](https://ai-weekly.ai/newsletter-05-26-2026/) --- ## 2. Project Glasswing: 10,000+ Critical Vulnerabilities Found in 30 Days On May 22, Anthropic published its first progress update on **Project Glasswing** — and the numbers are difficult to overstate. In approximately 30 days, Claude Mythos Preview and ~50 partner organizations identified **more than 10,000 high- or critical-severity vulnerabilities in systemically important software**. Of those, 6,202 were classified high or critical across 1,000+ open-source projects, with 1,094 confirmed as valid true positives. The most striking individual finds: a **27-year-old remote crash vulnerability in OpenBSD** — an OS literally built around security as its core value — and a **16-year-old flaw in FFmpeg**, the audio-video library embedded in YouTube, Netflix, Zoom, Chrome, VLC, and thousands of other applications. A critical **WolfSSL vulnerability (CVSS 9.1)** was also discovered — an embedded TLS library used in automotive systems, industrial controllers, and IoT devices. Partner disclosures are equally notable: **Cloudflare found 2,000 bugs** (400 high/critical) in its own systems; **Mozilla fixed 271 Firefox vulnerabilities**, a tenfold increase over its previous AI-assisted fix rate. The most important takeaway from the Glasswing update is not the 10,000 number. It is what it revealed: **the bottleneck in cybersecurity has shifted from finding vulnerabilities to fixing them**. AI can now generate valid vulnerability reports faster than human teams can verify, disclose, and patch them. 📖 [Read More — Build Fast with AI: May 30 AI News](https://www.buildfastwithai.com/blogs/ai-news-today-may-30-2026) --- ## 3. DeepSeek Makes Its 75% Price Cut Permanent — The AI Price Floor Just Collapsed On **May 22, 2026**, DeepSeek announced that the 75% promotional discount on its V4-Pro API — originally set to expire May 31 — is now the **permanent list price**. V4-Pro input tokens dropped from $1.74 to **$0.435 per million**; output tokens from $3.48 to **$0.87 per million**. Cache-hit pricing fell even further, to $0.003625 per million. The gap against Western frontier models is stark: V4-Pro is now roughly **11.5x cheaper than GPT-5.5 on input** and **34.5x cheaper on output**, at benchmark scores sitting within three to six months of the frontier. This is not a discount story — it is a structural repricing of what frontier-adjacent intelligence costs. DeepSeek explained that V4-Pro's new Compressed Sparse Attention architecture needs only 27% of the per-token inference compute and 10% of the KV-cache memory of its predecessor, making the cut permanent rather than promotional. For anyone building AI agents or high-volume LLM pipelines: a workflow running 100M output tokens monthly costs ~$87 on V4-Pro vs. ~$2,500 on GPT-5.5 or Claude Opus 4.7. That difference moves agent infrastructure from a line item to a budget conversation. 📖 [Read More — The Next Web: DeepSeek Price Cut](https://thenextweb.com/news/deepseek-v4-pro-75-percent-price-cut-permanent) | [Read More — InfoWorld: AI Pricing War](https://www.infoworld.com/article/4176709/deepseeks-steep-v4-pro-price-cut-escalates-ai-pricing-war.html) --- ## 4. Microsoft Copilot Studio: Computer-Using Agents Hit General Availability Microsoft published its **Copilot Studio May 2026 update** on May 26, with the headline that **computer-using agents are now generally available**. These agents can interact directly with desktop and web applications exactly as a human would — navigating screens, clicking, filling forms, and extracting data from any application with a visual interface, regardless of whether an API exists. This is the capability that makes Copilot Studio agents useful for the long tail of enterprise software: the 10 to 50 internal tools organizations run that were never designed for programmatic integration. Three other significant improvements ship alongside it: a **redesigned workflow canvas** with conditional branching, parallel execution paths, and step-by-step debugging; **Work IQ extensibility** allowing agents to read and act on signals from Microsoft Viva workforce analytics; and **real-time voice experiences** with sub-500ms response latency for customer service and helpdesk use cases. Microsoft is rebuilding Copilot from a conversational assistant into an agent-first multi-model platform — with Azure as the governance and security layer regardless of which underlying model is doing the reasoning. 📖 [Read More — Build Fast with AI: May 27 AI News](https://www.buildfastwithai.com/blogs/ai-news-today-may-27-2026) --- ## 5. Claude Opus 4.8 Released — 41-Day Release Cycle, Dynamic Workflows Anthropic released **Claude Opus 4.8** on **May 28, 2026**, just 41 days after Opus 4.7 — its fastest major model release cycle to date. The update introduces **Dynamic Workflows**, a capability allowing Claude to adaptively restructure its own multi-step task plans mid-execution based on intermediate results, rather than committing to a fixed plan at the start. Early analysis describes this as a meaningful step toward more resilient agentic behavior: agents that can recover from unexpected tool outputs, API failures, or information gaps without requiring human intervention to restart the task. The release arrives alongside IBM joining **Project Glasswing** — IBM Research is now applying Claude to identify and remediate vulnerabilities in widely used software, contributing fixes back to the open-source community via IBM Concert, its AI-driven vulnerability management platform. 📖 [Read More — AI-Weekly Issue 218](https://ai-weekly.ai/newsletter-05-26-2026/) --- ## 6. Anthropic Opens Milan and Seoul Offices in the Same 48-Hour Window Anthropic opened **two new international offices in a single 48-hour window** — Milan (May 27) and Seoul (May 26) — bringing its total international presence to six offices outside the US alongside London, Tokyo, Bengaluru, and Singapore. The Korea expansion is driven by a striking data point: Claude usage in South Korea runs **3.5x higher than population would predict**. KiYoung Choi — formerly Country Manager of Snowflake Korea, Google Cloud Korea, and COO of Microsoft Korea — has been appointed as Anthropic Korea's Representative Director. The Milan office serves Italian enterprise, research, and developer communities (manufacturing, fashion, financial services, and public sector) and creates a pathway for GDPR-compliant European access to Claude Security — currently unavailable to European organizations due to data sovereignty constraints. Meanwhile, Anthropic's Seoul opening includes active discussions with South Korea's **Ministry of Science and ICT, National Intelligence Service, Financial Services Commission, and National AI Strategy Committee** about Mythos cybersecurity cooperation — suggesting Korea could join the Glasswing consortium in some form in the near term. 📖 [Read More — Build Fast with AI: May 30 AI News](https://www.buildfastwithai.com/blogs/ai-news-today-may-30-2026) --- ## 7. SpaceX IPO: Roadshow Starts June 4, Pricing June 11, Trading June 12 The SpaceX IPO timeline is now locked. The investor roadshow begins **June 4**. **Pricing is June 11**. Trading opens on Nasdaq under the ticker **SPCX on June 12**. The offering targets a **$1.75 trillion valuation** and a raise of up to **$75 billion** — which would be the largest IPO in capital markets history, surpassing Saudi Aramco's 2019 $35.4B record. Goldman Sachs leads a 21-bank underwriting syndicate. In an unusual move, 30% of the float is being routed directly to retail investors through Robinhood, Fidelity, and Charles Schwab. Polymarket prediction markets are pricing a 94% probability of completion within the June window. One number the Anthropic ecosystem should watch: the **$1.25B/month compute deal Anthropic has with SpaceX** (through May 2029) is disclosed in the S-1, and could add ~$2.5B in quarterly SpaceX revenue as it ramps — shifting the IPO narrative materially at the first earnings call in September. 📖 [Read More — Build Fast with AI: May 27 AI News](https://www.buildfastwithai.com/blogs/ai-news-today-may-27-2026) --- ## 8. AI Outperforms Average Humans on Creativity Tests — 100,000-Person Study A peer-reviewed study comparing **more than 100,000 people** with today's most advanced generative AI systems found that AI now **outperforms the average human on standardized creativity tests** — including the Alternative Uses Task, the Remote Associates Test, and divergent thinking measures. The finding is statistically robust given the sample size, though researchers emphasize the tests measure a specific kind of structured creative output, not the open-ended, culturally embedded creativity behind great literature or music. Two practical implications stand out. First, these are the exact tests used in organizational psychology for decades to screen for creative potential in hiring — if AI outperforms the average hire on tests that predict creative job performance, the hiring implication is real. Second, the productivity gains from AI-assisted creative work may be largest in the middle of the distribution — the roles that require average, not exceptional, creativity are most exposed to automation first. 📖 [Read More — Build Fast with AI: May 27 AI News](https://www.buildfastwithai.com/blogs/ai-news-today-may-27-2026) --- ## 💼 Impact & Opportunities for Salesforce Consultants The week ending May 30 delivered a set of stories that are directly legible as forces reshaping enterprise technology consulting: a permanent collapse in AI inference pricing, a new generation of computer-using agents, and a stunning proof point that AI can identify security vulnerabilities at a scale and speed that no human team can match. Each of these has a concrete translation for Salesforce practitioners. ### ⚠️ Role Impact: How Your Work Is Changing - **DeepSeek's permanent price floor changes the build-vs-configure conversation.** When frontier-adjacent AI inference costs drop to $0.87/M output tokens, the economics of building custom AI capabilities shift dramatically. Tasks that previously required expensive Claude or GPT-5.5 API calls — document extraction, classification, summarization at scale — can now be routed to DeepSeek at a fraction of the cost. Clients will increasingly ask why their Salesforce AI implementations are priced the way they are, and consultants who can't answer that question with a coherent vendor tiering strategy will look behind the curve. - **Microsoft's computer-using agents are a direct competitor to integration work.** Copilot Studio's computer-using agents can now navigate any application with a visual interface — no API required. For Salesforce consultants who build custom integrations between Salesforce and legacy systems, this is the capability that has been waiting to eliminate a significant category of integration work. Screen-scraping, form-filling, and legacy ERP data extraction projects that take weeks to build are now one Copilot Studio agent configuration away from being solved by the client's IT team without an SI. - **The creativity study has a specific implication for consulting deliverables.** The jobs most exposed to AI displacement are those requiring average creativity — the kind involved in producing standard business requirements documents, process flow diagrams, data dictionaries, and solution design templates. If you are a consultant whose primary value is producing well-formatted documents rather than making judgment calls about architecture, governance, or organizational dynamics, the exposure is real and growing. The consultants best protected are those solving non-standard problems that require judgment, not pattern-matching. - **Glasswing's discovery-to-remediation bottleneck is coming to enterprise security engagements.** When Mozilla runs Claude Mythos and generates 271 patch-worthy vulnerabilities at 10x its previous rate, every enterprise CISO reading that story is immediately calculating how long their backlog of unpatched vulnerabilities has been sitting untouched. Salesforce orgs are no exception — every large enterprise has years of accumulated technical debt in their org that a Mythos-level scan would surface. Consultants who aren't ahead of the security conversation risk being caught flat-footed when clients start asking about AI-assisted org security audits. ### ✅ Opportunities: Where to Play Offense - **Build a cost optimization practice around AI model tiering.** DeepSeek's permanent price floor is the opening you need to have a genuine cost-efficiency conversation with clients about their Salesforce AI architecture. Not every agentic task needs Claude Opus 4.8. Map your clients' Agentforce workflows by task complexity and design a tiered model routing strategy — frontier models for high-stakes reasoning, mid-tier for standard processing, cost-optimized models for high-volume classification. Clients who implement this will cut AI infrastructure costs by 50–80% on applicable workloads, and you become the consultant who saved them real budget. - **Design Agentforce for computer-using agent compatibility.** Microsoft's computer-using agents reaching GA is a signal that this pattern — agents operating desktop and web UIs without an API — is about to become the standard approach for connecting Salesforce to the long tail of unintegrated enterprise systems. Rather than building every integration the traditional way, design Agentforce implementations with a hybrid architecture in mind: API-first where APIs exist, computer-using agents where they don't. This saves implementation time and future-proofs the design. - **Lead the Claude Opus 4.8 Dynamic Workflows conversation.** The Dynamic Workflows capability — agents that adaptively restructure their own task plans mid-execution — is a meaningful shift in what Agentforce-Claude integrations can do for complex, multi-step business processes. Consultants who can walk clients through concrete use cases (a multi-step customer onboarding process that recovers gracefully when a background check API times out, for example) will win the architecture conversations that turn into large implementation projects. - **Pitch AI-assisted Salesforce org security reviews before clients ask for them.** The Glasswing findings — 400 critical bugs at Cloudflare, 271 Firefox vulnerabilities at Mozilla — will prompt every enterprise security team to wonder what an equivalent scan would find in their Salesforce org. Consultants who already have a structured offering for AI-assisted Salesforce security assessments (permissions review, sharing model audit, API credential exposure, agentic workflow trust configuration) will be first in line when that conversation starts. Start building the methodology now. - **Use Copilot Studio computer-using agents to accelerate your own implementation delivery.** The same capability threatening integration project work can turbocharge your own. Computer-using agents can handle repetitive migration tasks, test data population, regression testing across Salesforce screens, and configuration verification — the work that consumes junior consultant hours during delivery. Consultants who build internal workflows using computer-using agents will have meaningfully lower delivery costs and can price more competitively while protecting margin. --- _Brief compiled May 30, 2026 | Sources: Build Fast with AI, The Next Web, InfoWorld, AI-Weekly_