OpenAI changed cyber rules. Is your team in the queue?

This week inside.:

It’s Official: AI Is Making Your Best People Even Better

A landmark field study from MIT Sloan found that GenAI boosted strong performers by over 20% and dragged weak ones down by nearly 10%. The gap between your AI winners and losers is not a coincidence. It is a data problem.

MIT Sloan Management Review just published research that should land on every enterprise AI leader's desk today. A five-month field experiment by researchers Nicholas Otis, Rowan Clarke, Solene Delecourt, David Holtz, and Rembrand Koning found that generative AI does not create uniform performance gains. It amplifies what is already there, for better or worse.

The Finding That Will Change How You Deploy AI Across Teams

The study ran 640 entrepreneurs through a randomized trial. One group received a standard business guide. The other received a GPT-4-powered AI business mentor via WhatsApp. The average treatment effect? Near zero. Not because AI did nothing, but because the gains and losses canceled each other out entirely.

• High performers: +20% revenue lift. Strong operators used AI to validate decisions, spot opportunities, and move faster

The root cause is not the AI. It is the underlying data literacy and decision quality of the person using it. Strong performers could evaluate AI outputs, push back on bad suggestions, and apply the advice selectively. Weak performers treated AI outputs as instructions and followed them off a cliff.

What This Means for Your Enterprise AI Rollout Right Now

If you are deploying AI tools across your organization without first assessing the data quality and decision-making baseline of each team, you are not democratizing productivity.

You are widening the performance gap between your best and worst operators. The same AI, the same prompt, the same output, can generate opposite business outcomes depending on who receives it.

• AI amplifies existing capability gaps rather than closing them across teams

• The quality of data feeding your AI systems directly determines which way that amplification goes

• Organizations without a baseline data management strategy will see AI deepen inequality in team output

• High performers benefit most when AI is connected to clean, governed, reliable data sources

The Fix Is Not a Better AI Tool. It Is Better Data Under It.

The MIT Sloan research makes one structural point unmistakably clear. The variance in AI outcomes tracks directly to the quality of the inputs and the judgment of the operator. For your enterprise, judgment develops over time. But data quality is something you can fix right now. Teams that win with AI are almost always teams that have already solved for clean, structured, and governed data pipelines before the AI layer ever touches them.

Your AI Is Only As Good As the Data Under It. Is Yours Ready?

The MIT Sloan study reveals that performance gaps come down to data quality and the decision-making baseline. DataManagement.AI gives your enterprise the governed, structured data foundation that turns AI from a liability into a force multiplier, before another team quarter is lost to bad outputs.

See how DataManagement.AI closes the performance gap

Your Security Team Is Already Behind - OpenAI Just Changed the Rules

GPT-5.4-Cyber is live, vetted defenders are getting access right now, and if your organization isn't in the queue, your adversaries may already be ahead of you.

• 3K+ Fixed critical vulnerabilities via Codex Security

• 76% Capture-the-flag benchmark score (up from 27%)

• 1K+ Open-source projects scanned for free

OpenAI has just released GPT-5.4-Cyber - a purpose-built, fine-tuned variant of its flagship GPT-5.4 model, and it's already being handed to verified security teams.

If you run a security operation, manage critical software infrastructure, or advise enterprise clients on cyber risk, this directly affects your competitive position today.

What exactly just landed on your threat landscape

Unlike the standard GPT-5.4, this model is explicitly "cyber-permissive." That means your verified analysts can now conduct binary reverse engineering - examining compiled software for malware and vulnerabilities without needing source code access - without the model refusing them mid-workflow. It's a material productivity shift for your red and blue teams.

Why you can't afford to sit this one out

Here's the uncomfortable truth: AI is a dual-use technology. The same capabilities that let your defenders move faster also exist for adversaries. OpenAI openly acknowledges this. The question isn't whether AI-augmented attacks are coming - they're already here. The question is whether your team has the same caliber of tools as your adversaries are building with.

The access model your procurement team needs to understand now

You don't get GPT-5.4-Cyber by simply upgrading your API tier. OpenAI is operating a tiered verification system: individual defenders verify identity through OpenAI's cyber access process; enterprise teams apply through their OpenAI account representative.

The highest tier unlocks the full cyber-permissive model. Start your verification process now - the rollout is gradual, and demand is already significant.

The AI Model Storage Problem Everyone Ignores Just Got a Free Fix.

Project Pipit compresses large language models up to 5.2x without touching a single number. Zero accuracy loss. Zero new pipelines. And it is completely open source.

• 5.2x Compression on dense Llama-3 class models

• 3.8x Compression on Mixture-of-Experts architectures

• 0 Accuracy loss. Byte-for-byte identical outputs

Cloudflare has open-sourced Project Pipit, a lossless compression tool that could fundamentally change how your enterprise stores, distributes, and runs large AI models. If you are currently paying GPU cloud rates to serve models you built, you need to understand what just landed.

Why Every Other Compression Tool Is Costing You Accuracy

Quantisation and pruning have long been the industry default. You shrink the model, you accept some degradation, and you move on. Pipit refuses that tradeoff. Built under Dr. Adaosa Okafor using a proprietary entropy-coding algorithm, it compresses models in a fully reversible way.

Your probability distributions, benchmark scores, and outputs remain identical after decompression. Your compliance benchmarks stay intact.

Pipit supports PyTorch and SafeTensor formats out of the box, meaning your team can adopt it without touching existing pipelines or retraining any models.

The Numbers That Will Change Your Infrastructure Conversation

For models exceeding 70 billion parameters, reduced network transfer times from Pipit outweigh any minimal runtime overhead. Integrated with Cloudflare Workers AI, compressed models stream from edge locations and decompress with near-zero latency. Storage and bandwidth costs drop by up to five times.

The Part Nobody Talks About: Moving Your Data Into These New Environments

Cheaper, edge-ready AI infrastructure sounds transformative until you factor in the complexity of getting your enterprise data there.

Migrating data to new cloud or on-premise environments where these compressed models will run is where most projects stall, run over budget, and lose accuracy. That gap between AI infrastructure savings and actual deployment is where enterprises bleed cost.

If widely adopted across the industry, Pipit has the potential to become the standard compression layer for AI model distribution and to accelerate decentralised AI deployments globally.

The enterprises that pair infrastructure savings from tools like Pipit with automated, AI-driven migration pipelines will be the ones that actually capture that cost reduction, not just read about it.

You Are Saving on AI Infrastructure. Now Save on the Migration Too.

DataManagement.AI eliminates the manual work that blows migration budgets. Profile AI discovers your data landscape automatically. Map AI handles schema transformations. Reconcile AIguarantees 100% post-migration accuracy. Your team ships faster, not harder.

What If AI Ran Your Work While You Slept? A $25M Startup Just Made It Real.

A Cupertino startup just closed its third funding round in under a year. The pitch is not a better chatbot. It is an AI that builds your tools and runs them while you sleep.

Most AI startups compete on model performance. Creao AI is pitching a better loop. The Cupertino-based startup just closed a $10 million round led by Prosperity7 Ventures, the diversified venturing arm of Aramco Ventures, bringing total funding to $25 million across three rounds in under a year. 

If your enterprise is still waiting for AI to actually do the work rather than just describe it, this round signals something your operations team needs to pay attention to right now.

The Productivity Ceiling Your Team Has Already Hit

Every AI tool your team uses today has the same silent problem. A human still has to press run. The gap between a chatbot answering a question and an agent actually running your work while you sleep is where enterprise productivity stalls. Creao AI's CEO Kai Cheng, who spent a decade building production AI systems for over 250 enterprise clients, put it plainly.

What a Closed-Loop AI System Actually Means for Your Operations

CREAO starts as a conversation. Users describe a task to what the company calls a "super agent," a cloud-based AI that not only answers questions but also executes. It writes code, calls APIs, connects integrations, and delivers results in a sandboxed environment. What happens next is the differentiator.

Successful work gets saved as a reusable Agent App, a persistent, schedulable unit of automation with its own memory. That SEO pipeline you ran on Tuesday runs again on Friday. Automatically. Without you. For your operations, finance, and marketing teams, this is the difference between AI as a productivity assistant and AI as a workforce multiplier.

Why Investors Are Betting $2 Billion on This Category Right Now

The capital flowing into autonomous AI agents is not speculative. Manus, the general-purpose AI agent that topped the GAIA benchmark, was acquired by Meta for $2 billion. 

Genspark hit unicorn status with a $275 million Series B at a $1.25 billion valuation, reaching $50 million in ARR just five months after launch. Gumloop closed a $50 million Series B from Benchmark for its no-code agent builder targeting enterprise automation.

• Meta paid for Manus $2B, Validating autonomous agents as platform-level infrastructure

• Gumloop Series B $50M From Benchmark, targeting enterprise no-code automation

The One Signal That Matters More Than the Funding

Creao AI has reached 200,000 organic users without paid marketing. For your enterprise evaluation checklist, that is the metric that matters. Product-led growth at this scale, in this category, tells you the underlying workflow problem is real and that the solution is resonating before any sales team gets involved. 

The funding will expand Creao's engineering team, deepen enterprise integrations, and scale agent-to-agent collaboration across its platform. 

Your competitors are already evaluating it. The question is whether you are.

Your opinion matters!

Hope you loved reading our piece of newsletter as much as we had fun writing it. 

Share your experience and feedback with us below ‘cause we take your critique very critically. 

Thank you for reading

-Shen & Towards AGI team