Grok Just Beat Google’s CAPTCHA 🤖

Advertise here | 6-min Read

Dumping a spreadsheet into Claude or ChatGPT gets you surface-level guesses, not answers. AI can actually diagnose what’s wrong with your marketing, but only if it can see complete and  connected marketing data. See how SegMetrics can help you turn your AI into a trustworthy analyst.

No more data hallucinations or vague summaries.

Grok Just Beat Google’s CAPTCHA 🤖

WHAT’S HAPPENING AI TODAY

1. Microsoft patched CoSnitch — a Copilot memory-poisoning vulnerability that auto-exfiltrated Gmail, Drive, and Outlook via one malicious link: Microsoft shipped a patch on Aug 18 for CVE-2026-24301, dubbed CoSnitch by Varonis Threat Labs. The flaw chained an undocumented URL parameter, Copilot's built-in URL fetch, and persistent memory poisoning so a single malicious link auto-executed prompts and exfiltrated connected Gmail, Drive, and Outlook data. CoSnitch is the most severe documented Copilot vulnerability to date — a single malicious link, no user interaction required beyond clicking it, and automatic exfiltration of every connected Google and Microsoft account.

2. NVIDIA is in talks to invest in Mercor at $20 billion — double its valuation from nine months ago: The Information reports NVIDIA is discussing an investment in Mercor as part of a round that would value the AI data-labeling startup at $20B, roughly double its $10B Series C nine months ago. NVIDIA paid Mercor tens of millions last quarter; Mercor's H1 gross revenue hit $614M with a $2B annualised run rate. NVIDIA investing in the company it already pays tens of millions per quarter for data labeling is vertical integration in the AI training stack. At a $20B valuation and $2B annualised revenue.

3. SpaceX is pursuing Cognition — the Devin AI coding agent company — after acquiring Cursor in August: SpaceX pursued Cognition, now weighing compute. Cognition built Devin — the first AI software engineer to pass real engineering interviews and complete real tasks on production codebases. SpaceX already owns Cursor (Anysphere), which it acquired for $60 billion in August, and Origin, Cursor's GitHub competitor. A Cognition acquisition would give SpaceX ownership of two of the top three AI coding agent platforms — Cursor for the IDE layer, Devin for the autonomous coding layer — plus Colossus compute infrastructure.

AI NEWS HIGHLIGHT

• OpenAI 2027 IPO committed — monitoring overhead now 20% of frontier inference compute — the first time any frontier lab has quantified its safety cost. Scales with usage. Not itemised in per-token pricing.
• CoSnitch patched — CVE-2026-24301 auto-exfiltrated Gmail, Drive, Outlook via one malicious link — verify the August 18 patch is applied. Audit Copilot memory for entries added by untrusted sources since August 1.
• NVIDIA investing in Mercor at $20B — data labeling now valued like AI itself — $2B annualised revenue, 10x multiple. NVIDIA is vertically integrating the training data stack it already pays for.
• SpaceX pursuing Cognition (Devin) after acquiring Cursor — developer tools consolidating around SpaceX — Cursor + Devin + Colossus compute. The developer infrastructure stack is consolidating faster than any platform since GitHub.
• Anthropic October IPO roadshow — two frontier labs going public within 12 months of each other — without precedent in tech. Both must explain safety cost structures to public investors. OpenAI just set the first public benchmark: 20%.
• Claude 3 Haiku shuts down on Vertex AI in 3 days — August 23 is Saturday — if you have not migrated Vertex AI pipelines from claude-3-haiku, do it today. This weekend will be too late for some teams.
• Gemini 3.7 Flash introductory pricing — benchmark it against Sonnet 5 before both windows close — two introductory pricing windows closing simultaneously. The September 1 routing decision is now 12 days away.
• Claude Sonnet 5: 12 days left at $2/$10 — standard $3/$15 from September 1. Haiku Vertex cutover Saturday. CoSnitch patched this week. Three infrastructure actions, one week.

Plan Your AI Token Budget Before the 20% Safety Overhead

Prompt: You are an AI infrastructure strategist. OpenAI just disclosed that monitoring overhead for GPT-5.6 Sol-class models and Astra now consumes roughly 20% of the inference compute being monitored. That cost is embedded in the per-token price — it does not appear as a line item, but it is real and it scales with usage. Claude Sonnet 5 introductory pricing ends in 12 days. Microsoft patched CoSnitch, a Copilot memory-poisoning vulnerability, on August 18. The Haiku Vertex shutdown is Saturday. Three infrastructure actions are converging in one week.

Here is our current AI usage: [describe your team size, which models and tools you use, estimated monthly token volume, monthly spend, and which workflows are highest priority].

Help me build an AI cost and governance framework across four areas:

1. The true cost model — OpenAI's 20% monitoring overhead means the effective cost of frontier inference is higher than the per-token price suggests. For our current usage: what is the implied total cost of our AI spend including safety overhead, estimated latency impact, and operational overhead for managing model migrations, deprecations, and security patches? Build the full cost picture, not just the invoice.

2. The September 1 routing decision — Sonnet 5 moves from $2/$10 to $3/$15 in 12 days. With CoSnitch patched, Microsoft Copilot's security posture has improved — but the incident raises questions about any AI tool with persistent memory and connected cloud accounts. For each of our current AI tools with connected account access: what is the minimum permission set required, and which connections can be revoked without losing meaningful functionality?

3. The Copilot memory audit — CoSnitch exploited Copilot's persistent memory. For any team using Microsoft Copilot: review your Copilot memory entries, identify any that were added by external sources or that reference data you did not explicitly share, and document the review. This is not a one-time action — memory poisoning attacks will recur. Design the monthly review process now.

4. The IPO signal read — OpenAI committing to a 2027 IPO and Anthropic's October roadshow are the most significant vendor stability signals of the year. For enterprise procurement: what does a frontier AI lab going public mean for contract terms, pricing stability, and the risk of vendor pivots? What contractual protections should we be negotiating now, before the roadshow creates pricing pressure?

TOP TRENDING AI TOOLS

• Hey Noah — AI assistant that remembers your work and proactively surfaces what you need
• Perplexity AI — AI-powered search with fact-checked answers and direct citations
• Claude — Thoughtful, context-heavy writing and deep document analysis
• Cursor — Repo-aware AI code editor — refactoring, debugging, PR support, terminal execution
• Kane CLI — AI terminal agent that executes tasks, runs tests, and explains errors
• Murmell — Cloud canvas where your team and AI agents work together

• AgentSky — AI agents for cross-app task execution, no-code setup
• Ito — AI that manages your calendar, inbox, and tasks proactively
• Wispr Flow — Dictate in any app — 4x faster than typing, writes in your voice, 100+ languages
• Soloop — AI video editor that auto-cuts, captions, and repurposes long-form content
• ElevenLabs — Industry standard for voice cloning, text-to-speech, and audio generation

That’s a Wrap

SPONSOR US

Get your business in front of over 90k+ AI professionals

8020AI is the world’s #1 AI Newsletter, Read by 90k+ professionals from leading companies such as Google, OpenAI, Meta, and Microsoft.

We've assisted in promoting Over 500 AI-Related Products. Will yours be the next?

What We Can Offer:

• Launch an Advertising Campaign

• Introduce New Product or Features

• Other Business Cooperation

Get in touch today.

Or Email our founder Alamin at [email protected]

FEEDBACK

How was your experience with 8020AI today?

Login or Subscribe to participate in polls.

If you have specific feedback or anything interesting you’d like to share, please let us know by replying to this email.