In partnership with
Welcome Automaters, 👋
The global AI race just delivered a top-tier cybersecurity scandal.
Here’s the exact tea: China’s National Vulnerability Database (NVDB), a government cybersecurity platform operated under the country's industry ministry, dropped a massive warning on its official WeChat channel. Their claim? Anthropic’s Claude Code has a "serious" security backdoor baked straight into its software.
According to the official NVDB advisory, researchers detected a hidden monitoring mechanism inside the software. They claim it quietly harvests sensitive developer data and transmits it to remote servers without asking for user consent.
Here’s what the watchdog says is being sent behind your back:
Geographic Location: Precise location details of where the code is being run.
Identity Identifiers: Specific user credentials and account mapping data.
Network Traffic: Internal system connections tied to core developer networks.
The warning specifically targets Claude Code versions 2.1.91 through 2.1.196, which covers releases pushed out between April and late June. So if your team is rocking one of those builds, China's advice is completely blunt: uninstall it or upgrade immediately.
Now, here's where the story gets really juicy. Anthropic is not panicking at all, and their engineers are publicly setting the record straight.
What Chinese regulators are dramatically calling a "backdoor" is actually an experimental anti-abuse safety measure. Anthropic engineers explained on X (formerly Twitter) that the code was deployed back in March to catch unauthorized proxy resellers and stop rival companies from "distilling" (aka reverse-engineering and stealing) Claude's secret sauce.
Plus, Anthropic points out something kind of important: Claude access is not even officially permitted in China in the first place!
This public callout did not happen in a vacuum. Just last week, Chinese tech giant Alibaba officially banned its employees from using Claude Code at work, instructing engineers to switch to their in-house tool, Qoder.
This came right after Anthropic publicly called out Alibaba for orchestrating a massive model distillation campaign, allegedly using roughly 25,000 fake accounts to siphon off Anthropic's proprietary logic.
The Bottom Line:
One side is yelling "backdoor," while the other is shouting "safety feature." Either way, this is proof that the global AI arms race has evolved into a full-blown trust war, and neither side is holding back.
To read the original legal and technical breakdown, check out the full Reuters exclusive report.
Here's what we have for you today
🤖 Grok 4.5 Release Explained: Price, Specs, and How It Compares to Opus (Plus Meta's New Anti-Spy Glasses Rule)
First up, Elon Musk’s newly rebranded SpaceXAI just officially unleashed Grok 4.5, marking the company's very first model release since going public.
Think of Grok 4.5 as the ultra-confident transfer student who struts into AP Physics announcing: "I can do everyone's homework in half the time, and I will charge you way less money to do it."
So, what exactly does this new model come with?
Built specifically for heavy coding, agentic office tasks, research, knowledge work and writing, Grok 4.5 is trained alongside Cursor data to handle serious real-world engineering loops.
Elon Musk went straight to X to pit Grok 4.5 directly against Anthropic’s powerhouse models, boasting that Grok 4.5 delivers performance comparable to Opus 4.7 and Opus 4.8, but runs significantly faster and costs a fraction of the price.
Here’s how the token pricing breaks down:
Grok 4.5 costs $2 per 1M input tokens, and $6 per 1M output tokens
Anthropic Opus 4.7: $5 per 1M input tokens, and $25 per 1M output tokens
OpenAI (GPT Tiers): $1 to $5 per 1M input tokens, and $6 to $30 per 1M output tokens
To make things even crazier, OpenAI is simultaneously rolling out GPT-5.6, its massive flagship update that was previously delayed over safety concerns. OpenAI has called it its “strongest model yet,” so yeah, it’s safe to say this week is packed!
Meanwhile, over on the wearable side of tech, Meta just decided to violently shut down an underground side hustle that has privacy advocates furious.
Image Credit: Meta’s Page
Meta’s newest AI smart glasses reignited intense public outrage over privacy. Critics raised alarms about people using the frames to secretly record bystanders, pointing out that modders had already figured out how to physically destroy or bypass the white "capture LED" light that blinks when recording.
In fact, an underground market popped up where modders were charging somewhere from $60 to $100 a pop to drill out the recording light while keeping the camera fully functional. Yikes.
Meta clapped back with an official privacy FAQ and a mandatory software update.
Here’s what is coming:
Auto-Shutdown: If the software detects that the LED is covered with tape or stickers, the camera immediately shuts off.
Hardware Anti-Tamper: The brand-new update detects if the LED circuit has been physically desoldered, drilled out, or destroyed. If it senses hardware damage, it completely disables the camera.
Legal Warfare: Meta is actively scrubbing marketplace listings for LED tampering services, banning offender accounts, and threatening to sue modders into oblivion, even if those ads run somewhere other than Meta's own platforms.
So essentially, that’s Meta saying: You're warned!
For the full technical breakdowns, check out the original reporting on TechCrunch and Engadget.
You Don't Have to Guess When, Only Which Way
That's a perp. No expiry, no rollover fees, no options complexity. Just a straight trade on direction, held as long as you want it. $90T trades on this market offshore every year. Kalshi just made it the first legal, CFTC-regulated version in the US.
Using leverage increases risk of loss. Leverage is subject to the Firm's review and the customer's risk profile.
🧱 Around The AI Block
🤑 The AI startups growing revenue at faster and faster rates.
🙅 The U.S. wants to restrict corporate use of Chinese AI.
🥳 OpenAI gets permission to roll out GPT-5.6 to the public on July 9.
🗣️ OpenAI releases new voice models for more natural live conversations.
🦾 Claude Cowork outperforms Gemini on email research tasks.
🛠️ Google Photos adds a new AI ‘Video Remix’ tool.
🦿 Mistral launches first robotics model in physical AI push.
🛠️ Trending Tools
For Curated Generative Talent: The Promptists is a specialized freelance marketplace and guild connecting brands with vetted AI prompt engineers and visual artists. It skips open-bidding noise by using an algorithmic matching engine to align creative briefs with a select group of practitioners based on style, domain, and rate parameters, complete with Stripe milestone protections.
For High-Volume SKU Generation: PixBulk is a programmatic image creation utility built for multi-channel e-commerce sellers on Shopify, Amazon, and Etsy. By ingesting a standard CSV or spreadsheet containing product SKUs and descriptive variables, it generates hundreds of studio-quality product lifestyle photos simultaneously, bypassing manual prompt input.
For Adaptive Email Voice Matching: ForthWrite is an AI email extension for Gmail and Outlook that drafts context-aware responses directly inside your inbox. It passively analyzes historical sent-email patterns to mirror your tone, sentence structure, and signature sign-offs, continually refining its writing style over time.
For Scientifically Accurate Diagrams: Figviz is an academic AI diagram engine engineered specifically for STEM educators and researchers. It converts plain-English prompts or complex PDF reference papers into labeled, high-contrast scientific illustrations, exporting scalable SVGs and PNGs optimized for research papers, textbooks, and lectures without text smudging or structural errors.
🤖 AI Workout Of The Day: Come Up With LinkedIn Post for Sharing Expertise Using AI
Sharing your real-world insights on LinkedIn does more than just fill a content calendar; it builds an algorithmic engine for inbound opportunity.
When you consistently translate technical or complex ideas into clear, actionable posts, you bypass traditional networking. You position yourself as a go-to practitioner in your field, which naturally attracts strategic partners, high-value clients, speaking invites, and career opportunities.
However, generic advice gets ignored. To break through the feed, your content needs a compelling hook, a scannable format, and a distinct, authoritative voice that invites conversation.
💡 Prompts to try:
Act as an elite B2B content strategist and LinkedIn personal branding specialist.
I want you to write a high-converting, authority-building LinkedIn post based on my insights about: [INSERT YOUR TOPIC OR CORE LESSON HERE]
To ensure the post stands out in the feed, please follow this proven structural framework:
1. THE HOOK (Line 1): Write a bold, pattern-interrupting opening line (under 12 words) that addresses a counterintuitive truth, a common industry myth, or a hard-learned lesson. Make it impossible not to click "see more."
2. THE FRICTION (Lines 2-3): Briefly call out the core problem or challenge most professionals face regarding this topic.
3. THE INSIGHT & ACTION (3-4 bullet points): Deliver the meat of the post using short, punchy bullet points. Focus on tactical, "how-to" value or an actionable mental model. Every bullet point should offer an immediate takeaway.
4. THE TAKEAWAY / MIC DROP: Synthesize the lesson into a single memorable sentence or rule of thumb.
5. THE CALL TO ACTION (CTA): End with a low-friction question designed to spark genuine debate or discussion in the comments (e.g., "What's your take on this?" or "How are you handling X in your team?").
STRICT FORMATTING & STYLE RULES:
* Length: 150 – 220 words total.
* Readability: Use single line breaks between thoughts. Keep sentences short and crisp. Optimize for mobile skimmability.
* Tone: Authoritative, conversational, and direct. Avoid corporate fluff, jargon, and empty buzzwords (like "game-changer," "synergy," or "delve").
* Hashtags: Include 3 to 4 hyper-relevant, targeted hashtags at the very bottom (blend broad industry terms with niche topics). Do not overuse them.
Optional Context to Include: [INSERT ANY SPECIFIC METRICS, PERSONAL ANECDOTES, OR LESSONS YOU WANT FEATURED]Is this your AI Workout of the Week (WoW)? Cast your vote!
That's all we've got for you today.
Did you like today's content? We'd love to hear from you! Please share your thoughts on our content below👇
What'd you think of today's email?
Your feedback means a lot to us and helps improve the quality of our newsletter.
