Top 5 AI News: June 2, 2026 - GitHub Kills the Flat-Rate Era for Copilot
The all-you-can-eat era for AI coding tools officially ended yesterday. GitHub switched its entire Copilot user base to token-metered billing on June 1, 2026 — and the ripple effects on how developers think about AI tools may be bigger than the billing change itself. That's the headline, but it's not the only story worth your attention today
1. GitHub Copilot Goes Metered: The Flat-Rate Era Is Over
Starting June 1, 2026, GitHub Copilot replaced its flat Premium Request Unit system with token-based "GitHub AI Credits," where one credit equals $0.01, and heavy usage — especially agentic coding sessions — can push users past their monthly allotment for the first time.
This is not a price increase. GitHub has been clear about that, repeatedly. What it is, is a structural shift in how developers experience AI tooling. Under the old model, a five-second autocomplete and a two-hour multi-file agentic session both cost the same: one premium request. GitHub was eating the difference. According to GitHub's chief product officer Mario Rodriguez, a quick chat question and a multi-hour autonomous coding session could cost the user the same amount — and the company couldn't sustain that as model complexity scaled.
The new system ties billing directly to token consumption across inputs, outputs, and cached context. Inline code completions and Next Edit Suggestions stay free forever — that's an important detail that got lost in the noise. What gets metered is everything agentic: chat, cloud agent loops, Copilot code review on pull requests. Individual Pro users get a monthly allotment of 1,000 credits ($10 worth). Pro+ users get 3,900 credits ($39 worth). Users can set a hard spending cap, making a runaway bill mathematically impossible if configured correctly.
My honest take: this was inevitable, and I'd argue it's actually good for the long-term health of AI tooling. The gym-membership pricing model worked when AI assistance meant autocomplete. It doesn't survive agentic workflows where a single session can consume thousands of tokens. The developers upset about this are often the power users who were getting the most subsidized value. For light users, nothing changes in practice.
The contrarian angle worth considering: this change might accelerate adoption of open-source alternatives like Ollama-based local models for routine tasks, with paid tools reserved for complex sessions. Developers will now have a reason to think carefully about when to invoke a cloud AI versus a local one — and that's a healthier mental model anyway.
Example 1 — Writing a Prompt That Minimizes Token Usage
You are a precise coding assistant. Answer only the specific question asked. Do not explain concepts I have not asked about. Do not add caveats unless there is a genuine bug risk. Output only: the code change needed, and a one-sentence explanation of why. My question: [specific coding question here]
Example 2 — Evaluating Whether a Task Is Worth Running as an Agent
I am deciding whether to use an AI agent or a simple chat prompt for the following task. Analyze it and tell me: (1) Is this task decomposable into steps that require multiple model calls? (2) Would a single well-structured prompt likely suffice? (3) What is the estimated token cost tradeoff? Task description: [describe your task here]
2. Iran Is Using ChatGPT and Gemini to Build Malware
Western AI models, including ChatGPT and Gemini, are actively being used by Iranian state-linked hackers to write malicious code, craft phishing emails in flawless Hebrew and Arabic, and scan infrastructure for vulnerabilities, according to an investigation published by the Financial Times on May 31, 2026.
This is the story I have been expecting for over a year. The progression was always obvious: as frontier models became powerful enough to write production-quality code, they would be exploited for offensive cyber operations. The FT's investigation confirms what cybersecurity analysts have been flagging in private. One anonymous analyst told the paper that Iranian operators appear to be using AI prompts end-to-end — not just for isolated tasks but as a full workflow accelerator.
The context matters here. Iran has been in a fragile ceasefire with Israel and the United States since early April 2026, but digital pressure has continued. Iranian hackers have reportedly used AI to generate ransomware-style "wiper" malware capable of destroying databases including backups. The UAE's Head of Cybersecurity, Dr. Mohamed Al Kuwaiti, previously reported that daily cyberattacks on UAE infrastructure doubled to over 500,000 incidents per day during recent escalation, with AI-written phishing emails described as flawlessly written in ways that used to be an easy red flag.
Here is the contrarian point everyone is dancing around: Iran doesn't actually need access to ChatGPT or Gemini specifically. Security experts like Leeron Walter at Teramind have noted that open-weight models like Meta's Llama or China's DeepSeek can be downloaded, run locally with no internet connection, and fine-tuned with zero guardrails. For a sanctioned nation-state, that is actually better operational security than using a monitored commercial platform. The FT story may be highlighting the tip of an iceberg that runs much deeper through open-source model abuse.
What this means for anyone building AI systems: prompt injection, jailbreaking, and misuse are not theoretical risks. Understanding system prompts and how guardrails work in commercial models is now baseline literacy for anyone thinking seriously about AI security.
Example 1 — Researching AI Security Threats for a Briefing
You are a cybersecurity analyst preparing a briefing for a non-technical executive. Explain, in plain language: (1) How can AI models like ChatGPT be misused for cyberattacks? (2) What specific attack types benefit most from AI assistance? (3) What defensive measures reduce AI-assisted attack success rates? Output format: executive summary (3 sentences), then bullet-point detail per section. Do not use jargon without defining it.
Example 2 — Analyzing Whether a Phishing Email Was AI-Generated
I need to analyze whether the following email was likely written with AI assistance for a phishing attempt. Check for: (1) Unusual fluency compared to typical phishing, (2) Specific personalization that suggests data reconnaissance, (3) Grammar and tone that is unusually polished for the claimed sender context. Email text: [paste email here] Output: likelihood score (1-10), reasoning, and recommended action.
3. China's Vast Raises ~$200M to Become AI 3D Modeling Unicorn
Beijing-based Vast, whose Tripo AI platform converts text and image prompts into production-grade 3D assets, has raised approximately $200M in cumulative funding and crossed the $1 billion unicorn valuation threshold, with its latest round led by Ince Capital and a venture fund backed by China Life Insurance Co.
The company was founded as recently as 2023 by 29-year-old Simon Song, who previously co-founded MiniMax, another Chinese AI unicorn. The speed of Vast's trajectory is striking even by Chinese AI standards: a $50M Series A in March 2026 alone, led by Alibaba Group and Hengxu Capital with participation from Baidu Ventures, was enough to push valuation past $1 billion. The Tripo platform claims 20 million global users and roughly 90,000 studios and companies, including major clients like NetEase and Sony.
I find this story more interesting than typical funding news because 3D asset generation is genuinely hard. Text-to-image generation had a clear quality ceiling before models pushed past it. Text-to-3D faces an additional challenge: mesh quality, topology, and production readiness for game engines and VFX pipelines. The fact that Sony and NetEase are active clients suggests Vast's output has cleared at least some bar for studio use. That's not a given in this space.
The broader context: this is part of a wave of Chinese AI companies operating in spaces where US export controls haven't yet created a meaningful capability gap. 3D modeling doesn't require the same compute density as training foundation models — it's applied AI that benefits from a large creative user base for training data. China's government investment in AI, including Vast's early backing from the Beijing Artificial Intelligence Industry Investment Fund, creates a funding floor for domestic AI startups that Western counterparts don't always enjoy.
Example 1 — Generating a 3D Asset Brief from a Text Description
I need to create a 3D asset brief for a fantasy game environment piece. Based on this description, generate: (1) A detailed text prompt optimized for AI 3D generation tools like Tripo or similar, (2) A list of style references to guide aesthetic direction, (3) Technical specifications (polygon budget, texture resolution, intended engine) I should specify to the tool. Asset description: [describe your 3D asset concept here]
Example 2 — Comparing AI 3D Tools for a Studio Workflow
I manage 3D asset production for a mid-size game studio. Compare the following AI 3D generation approaches for our workflow: - Text-to-3D tools (Tripo/Vast, etc.) - Image-to-3D tools - Traditional artist + AI-assisted retopology For each, assess: quality ceiling for AAA standards, speed advantage, integration with Unreal Engine 5, and current limitations. Recommend which approach suits different asset categories (hero assets vs. environment props vs. UI).
4. Daloopa Raises $47M to Fix AI's Financial Data Problem
Daloopa, a New York-based startup providing structured, source-linked financial data infrastructure for AI and agentic workflows in investment firms, raised a $47 million Series C on May 28, 2026, led by Brighton Park Capital, pushing the company's total funding past $100 million.
The company's core thesis is simple but important: the bottleneck in AI-powered finance isn't model intelligence, it's data quality. Daloopa's platform covers 5,500+ public companies globally, delivers up to 10 times more data points per company than competing providers, and hyperlinks every data point back to its original source filing for full auditability. CEO Thomas Li put it plainly: it's no longer enough for models to simply generate answers — they must be accurate and fully traceable.
This is the story underneath the story in enterprise AI adoption right now. Every large financial institution I follow has moved from "AI is interesting" to "AI is operational infrastructure" in the past 12 months. JPMorgan Chase formalized this in early 2026, reclassifying its AI investments as core infrastructure with a $19.8 billion technology budget and 2,000 staff dedicated to AI development. At that scale, a wrong number in an earnings model because the AI pulled inconsistent data from different fiscal calendars is a liability, not a minor inconvenience.
Daloopa's MCP (Model Context Protocol) integrations with ChatGPT, Claude, Perplexity, and Rogo are what make this genuinely interesting from a prompting perspective. Rather than a financial analyst asking an AI what a company's Q3 gross margin was and hoping the model scraped something accurate from the web, Daloopa functions as a verified data layer the AI draws from. The accuracy claim of greater than 99% and 14 years of historical data per company is what justifies the infrastructure framing.
The contrarian take: the bigger the role structured data infrastructure plays in AI outputs, the more the competitive moat in financial AI shifts away from whoever has the best model and toward whoever controls the most accurate data layer. Bloomberg and FactSet should be watching this closely.
Example 1 — Building a Financial Analysis Prompt Using Verified Data
You are a senior equity analyst at a value investment fund. Using the following structured data points (sourced and verified from company filings), build a comprehensive investment thesis for [Company Name]. Data provided: [paste your Daloopa or structured financial data here] Your output should include: (1) Revenue growth trend analysis over 5 years, (2) Margin expansion/contraction narrative, (3) Key risk factors from filing language, (4) A one-paragraph investment thesis with a 12-month outlook. Cite the specific data points that drive each conclusion.
Example 2 — Automating Earnings Update Workflow
I am an analyst who updates financial models every earnings season. Design a step-by-step agentic workflow prompt chain for processing a new earnings release. The workflow should: (1) Extract key metrics from the press release, (2) Compare actuals to consensus estimates, (3) Flag any guidance changes with specific numbers, (4) Identify 3 questions management will likely face on the earnings call, (5) Summarize the net impact on our existing model assumptions. Input: I will paste the earnings press release text. Output: structured sections matching steps 1-5 above.
5. Erin Brockovich Is Now Watching Your Data Center
Environmental activist Erin Brockovich — famous for her work against Pacific Gas & Electric and played by Julia Roberts in the 2000 film — launched a crowdsourced data center tracking map at brockovichdatacenter.com, receiving nearly 4,000 community submissions within the first month and more than 2,700 verified reports from across the US as of late May 2026.
This one caught me off guard, not because the issue is new, but because of who is now involved. Brockovich's name signals something specific in American public consciousness: this is the pattern that precedes class-action litigation. When she put out a call for submissions in April 2026, the response was immediate — over 1,600 reports in the first week alone, from 47 states. The biggest concentration came from Texas, specifically Sulfur Springs, where MSB Global is developing one of the largest AI data center projects on the continent: 3 gigawatts of capacity across 30 buildings on approximately 1,600 acres.
Brockovich has been careful to say she's not opposed to data centers or AI categorically. What she's opposing is the pattern: projects announced after permits are already secured, developers who don't return calls, local officials who signed NDAs before their neighbors knew a project was being considered. Eight percent of survey respondents reported that non-disclosure agreements were used to hide projects from the public until it was too late to object. Local utility rate increases driven by data center power demand are showing up as a recurring theme in the reports.
My take: the AI industry has been lucky that most public discourse has stayed at the model and application layer. The moment it moves to the infrastructure and utility layer — which is exactly what Brockovich's map is designed to do — the politics get significantly more complicated. A single hyperscale data center can consume as much electricity as a small city. Water usage for cooling is another major issue, especially in drought-prone Western states. These are real tradeoffs, not hypothetical ones.
Example 1 — Summarizing Community Impact Data for a Policy Briefing
I have compiled community reports about a proposed AI data center in [location]. Summarize these reports into a structured briefing for a city council member who is non-technical. Include: (1) Top 3 community concerns with frequency counts, (2) Specific utility and water impact claims with any numbers cited, (3) Comparison to typical data center environmental disclosures, (4) Questions the council member should ask the developer at the next public meeting. Community reports text: [paste reports here]
Example 2 — Evaluating AI Infrastructure Tradeoffs for an Article
I am writing a piece on the environmental and community impact of AI data center expansion in rural America. Help me structure a balanced analysis covering: (1) The legitimate economic benefits to host communities (jobs, tax revenue), (2) The documented environmental costs (power draw, water, land use, e-waste), (3) The transparency and process concerns raised by advocacy groups, (4) What responsible siting looks like based on current best practices. Tone: objective, specific, no boosterism for either side. Flag which claims need source verification.
Recommended Blogs
If you found this useful, these posts go deeper on related topics:
• Best Claude AI Prompts 2026: 25+ Types With Examples
• Best ChatGPT Prompts 2026: 200+ With Real Examples
• The Guide to Agentic Prompts
Frequently Asked Questions
What is the new GitHub Copilot billing system starting June 1, 2026?
GitHub replaced its Premium Request Unit system with GitHub AI Credits on June 1, 2026. One AI Credit equals $0.01 USD. Plan prices remain unchanged, but usage is now metered by token consumption. Pro users receive 1,000 monthly credits, Pro+ users receive 3,900, Business users receive 1,900 per seat, and Enterprise users receive 3,900 per seat. Inline code completions remain free on all plans.
Are Iran's cyberattacks using ChatGPT actually effective?
According to a Financial Times investigation published May 31, 2026, cybersecurity experts say Iranian hackers are using ChatGPT, Gemini, and other Western AI tools to write malware, craft phishing emails in flawless Hebrew and Arabic, and scan for infrastructure vulnerabilities at scale. One analyst described AI as helping Iran "absolutely raise their game." Security researchers also note that Iran likely uses open-weight models like Meta's Llama locally, which have no monitoring or guardrails.
What does Vast's Tripo AI actually do?
Vast's Tripo AI platform converts text and image prompts into detailed 3D objects and environments. The company, founded in 2023 in Beijing, reports 20 million global users and roughly 90,000 studios and companies on the platform, including NetEase and Sony as enterprise clients. The product competes in the text-to-3D space against tools from companies like Luma AI and Autodesk-backed World Labs.
Why did Daloopa raise $47M and what problem does it solve?
Daloopa raised a $47M Series C led by Brighton Park Capital in May 2026 to build structured, source-linked financial data infrastructure for AI-powered investment workflows. The core problem: most AI models hallucinate or pull inconsistent financial data from the web. Daloopa covers 5,500+ public companies globally with data hyperlinked to original source filings, achieving over 99% accuracy, and integrates directly with AI tools including Claude and ChatGPT via Model Context Protocol connectors.
What is Erin Brockovich's AI data center map and why does it matter?
Brockovich launched brockovichdatacenter.com in late April 2026, a crowdsourced map where US residents can report concerns about AI data center projects in their communities. As of late May 2026, the site had received over 2,700 verified reports from 47 states. Key concerns include utility rate increases, water usage, and non-disclosure agreements used to hide projects from the public before permits were secured. The initiative is widely seen as a potential precursor to class-action litigation.
How does GitHub Copilot's new metered billing affect heavy users?
Developers who run agentic coding workflows — multi-step agent loops, large-context code reviews, and extended chat sessions — are most affected by metered billing. Under the old flat-rate system, these power users were heavily subsidized. Under the new system, a complex agentic session consuming large token volumes will draw down credits faster than routine chat. GitHub allows users to set a hard spending cap. Inline autocomplete remains entirely free and does not consume credits.
Which AI models are most vulnerable to state-sponsored misuse?
Commercial models including ChatGPT, Google Gemini, and Claude are subject to terms of service enforcement and usage monitoring, which provides some deterrence. However, open-weight models like Meta's Llama 3 and DeepSeek R1 can be downloaded, run locally without internet access, and fine-tuned with no restrictions, making them a more operationally secure choice for state actors under sanctions. Security researchers consider locally-deployed open-weight models the greater long-term risk.
How should developers track their GitHub Copilot AI Credits usage?
GitHub has launched a preview billing dashboard accessible via the Billing Overview page on github.com. Users can monitor projected token consumption before the billing cycle closes, set per-organization spending caps, and review usage broken down by feature type. GitHub recommends that teams running agentic workflows review usage during the first billing cycle in June 2026 to calibrate expected monthly costs.
Follow along on promptailearning.com for weekly guides on prompting, AI tools, and getting more out of every model - news digest every Monday
References
1. GitHub Blog — GitHub Copilot Is Moving to Usage-Based Billing (Official Announcement, April 2026)
2. TechCrunch — Erin Brockovich Takes Aim at Data Center Secrecy (May 31, 2026)
3. Bloomberg — Gen-Z Gamer's 3D-Model Startup Vast Becomes China's Latest AI Unicorn (June 1, 2026)
4. PR Newswire — Daloopa Raises $47 Million Series C (May 28, 2026)
6. Brockovich Data Center Reporting — U.S. AI Data Center Awareness and Issue Map
7. The Hacker News — 2026: The Year of AI-Assisted Attacks (May 2026)