MIT Study Reveals A Staggering 95% Failure Rate For Generative AI Pilots

Here is what’s new in the AI world.

Explore Gen Matrix Q2 2025

Uncover the latest rankings, insights, and real-world success stories in Generative AI adoption across industries.

See which organizations, startups, and innovators are leading the AI revolution and learn how measurable outcomes are reshaping business strategies.

95% of Corporate AI Projects Are Failing, MIT Report Finds

A new, comprehensive report from MIT’s NANDA initiative, The GenAI Divide: State of AI in Business 2025, reveals a stark reality check for corporate America: despite massive investment and hype, the vast majority of enterprise generative AI projects are failing to deliver measurable financial results.

The 5% Success Gap

The research, based on 150 leader interviews, a survey of 350 employees, and an analysis of 300 public AI deployments, found that only about 5% of AI pilot programs achieve rapid revenue acceleration.

The other 95% remain stuck in pilot purgatory, stalling and delivering little to no impact on the company's profit and loss statement.

According to lead author Aditya Challapally, the divide is clear. Successful startups, often led by young founders, excel by focusing on a single pain point and forming smart partnerships, sometimes jumping from $0 to $20 million in revenue in a year. In contrast, most large enterprises are struggling to get their initiatives off the ground.

The Real Problem Isn't the AI

The core issue isn’t the quality of the AI models themselves. MIT’s research points to a major “learning gap” within organizations. While executives often blame regulation or technical performance, the true failure lies in flawed integration.

Tools like ChatGPT work for individuals because of their flexibility, but they stall in an enterprise setting because they cannot learn from or adapt to specific company workflows. Companies are trying to force a square peg into a round hole.

Misaligned Budgets and Successful Strategies

The report also uncovered a critical misalignment in spending. More than half of generative AI budgets are devoted to sales and marketing tools.

However, the biggest return on investment (ROI) was actually found in back-office automation, eliminating business process outsourcing, cutting external agency costs, and streamlining internal operations.

Furthermore, the strategy for adoption is crucial.

The data shows that -

Workforce Shifts and the "Shadow AI" Problem

The impact on the workforce is already underway, though not necessarily as mass layoffs. Companies are increasingly not backfilling roles in customer support and administrative positions as they become vacant, particularly those previously outsourced.

The report also highlights the persistent challenge of “shadow AI”, the unsanctioned use of tools like ChatGPT by employees, which creates security and compliance risks while making it difficult for leadership to accurately measure AI’s true impact on productivity.

The most advanced organizations are already moving beyond basic chatbots and experimenting with agentic AI systems, AI that can learn, remember, and act independently within set boundaries.

This shift offers a glimpse into the next phase of enterprise AI, where the technology becomes a truly autonomous driver of business processes.

Sam Altman Warns U.S. Is Underestimating China's AI Threat

OpenAI CEO Sam Altman cautioned that the U.S. risks downplaying the speed and seriousness of China’s advances in artificial intelligence and argued that export restrictions alone are unlikely to be an effective defense.

“I’m worried about China,” he said during a rare on-record discussion with a small group of reporters over lunch in San Francisco’s Presidio, not far from OpenAI’s first office.

Altman stressed that the U.S.-China AI rivalry is far more complex than simply tallying who is ahead. “There’s inference capacity, where China can probably scale faster. There’s research, there’s product. There are layers to all of this,” he explained. “I don’t think it will be as simple as: is the U.S. or China ahead?”

Despite Washington’s tightening of semiconductor export controls, Altman doubts that the measures are keeping pace with reality.

When asked whether restricting GPU shipments to China could be reassuring, he was skeptical: “My instinct is that doesn’t work. You can restrict one thing, but maybe not the right thing. People build fabs or find workarounds.”

U.S. policy has grown increasingly patchwork. After the Biden administration placed restrictions on advanced chip exports, former President Trump escalated the move earlier this year, blocking even chips designed to comply with earlier rules.

Yet just last week, regulators carved out an exception: allowing “China-safe” chips to be sold, provided Nvidia and AMD hand over 15% of their China revenue to the government.

Meanwhile, Chinese firms like Huawei are pressing ahead with homegrown alternatives, raising doubts over whether U.S. restrictions are really slowing progress.

China’s influence on OpenAI’s openness

China’s rapid AI development has also shaped OpenAI’s strategy on openness. Although the company has historically avoided fully open-sourcing its models, Altman admitted that competition from Chinese labs, especially open-source players like DeepSeek, factored into OpenAI’s recent decision to release its first open-weight models since 2019.

“It was clear that if we didn’t move, the world would be built mostly on Chinese open-source models,” Altman said.

Earlier this month, OpenAI launched two text-only open-weight models, gpt-oss-120b and gpt-oss-20b, intended as affordable options that developers can run locally and customize.

Unlike full open-source releases, the models’ training data and source code remain closed, but their parameters are freely available.

This shift marks a strategic turn: while Meta is reconsidering its openness with Llama, OpenAI is leaning into accessibility to expand its developer ecosystem and strengthen its edge against Chinese competitors.

Altman even admitted the company may have been “on the wrong side of history” by keeping its models closed for so long.

Reactions have been mixed. Some developers dismissed the models as stripped-down versions lacking many of the strengths of OpenAI’s commercial offerings. Altman acknowledged that the team optimized them primarily for one purpose: locally run coding agents.

20% of Blue-Collar Workforce Adopts AI Tools, Study Finds

India’s workforce is on the brink of a major transformation in how jobs are performed. According to The Work Ahead Report by job platform Indeed, professionals across the country are not only aware of artificial intelligence but are actively preparing to weave it into their career paths.

The survey, which included over 3,000 employees across both white- and blue-collar roles, reveals that 43% feel confident about adopting technologies like Generative AI and the emerging class of Agentic AI tools within the next two to five years.

For many, AI is no longer just a helpful add-on; it’s becoming a core skill tied to promotions, better pay, and long-term job security.

Mid-career workers are most confident

Confidence is especially strong among professionals aged 35 to 54, with nearly half (49%) saying they are ready for AI-driven workplaces. Surprisingly, this group feels more prepared than younger workers aged 18 to 24.

However, readiness comes with a demand: 56% of mid-career workers want more structured training to stay future-proof, compared with 41% of their younger counterparts. Their focus is practical, boosting efficiency, staying relevant, and advancing in their careers.

Yet optimism is tempered with caution. About one in three respondents worry about losing job security if they fail to keep up with technological shifts.

AI is becoming part of daily work

The report highlights that AI has moved beyond buzzwords into daily operations. One-third of professionals expect to use Generative AI regularly, while a quarter are preparing to adopt Agentic AI, systems that can handle complex tasks with little human oversight.

Even among blue-collar workers, the impact is clear. Seven in ten say technology already makes their jobs easier, and two in ten are actively using AI tools. From automating paperwork to improving customer interactions, AI is seeping into roles that were once untouched by digital disruption.

What employees expect

Workers are also asking for concrete support. About 29% want self-paced online programs to sharpen AI skills, while others prefer companies to allocate time during work hours for formal upskilling.

“There’s a growing confidence across India’s workforce,” said Sashi Kumar, Head of Sales, Indeed India. “Those who invest in AI skills will stand out for higher pay, promotions, and future opportunities. The rising interest in Agentic AI shows that people want to lead the change, not just react to it.”

The study makes one thing clear: in India’s job market, AI is no longer optional. It’s becoming a baseline skill essential for employability and career growth.

Open-Source Project 'Buttercup' Brings AI-Powered Patching

Buttercup is a free, AI-driven platform that automatically detects and fixes vulnerabilities in open-source software. Built by Trail of Bits, it recently secured second place in DARPA’s AI Cyber Challenge (AIxCC).

Open-source AI vulnerability scanner

Buttercup consists of four core components, each handling a different part of the process -

Requirements and setup

To run Buttercup, you’ll need at least an 8-core CPU, 16 GB RAM, 100 GB storage, and a stable internet connection for downloading dependencies.

The system relies on external AI models from OpenAI, Anthropic, and Google, which generate usage costs. A built-in budget control helps manage expenses.

Buttercup is open-source and freely available on GitHub.

OpenAI Launches Affordable ChatGPT Plan in India, Priced Under $5

OpenAI has introduced a new, budget-friendly ChatGPT subscription in India called ChatGPT Go, priced at ₹399 per month ($4.60). This makes it significantly cheaper than the existing ChatGPT Plus plan, which costs ₹1,999 per month (around $23).

The launch follows OpenAI’s recent shift to local currency pricing, and with Go, the company is also enabling payments via UPI, India’s popular digital payments system.

According to Nick Turley, VP at OpenAI and head of ChatGPT, the Go plan expands usage limits by up to 10x compared to the free tier, covering messages, image generation, and file uploads, while also enhancing memory features for more personalized responses.

Previously, the Plus plan effectively cost more than $20 when converted into INR, making it pricier for Indian users. The new Go plan offers a lower-cost alternative for people primarily using ChatGPT for conversations, image creation, and file handling.

Rumors of this plan first surfaced through Tibor Blaho, a software engineer known for credible AI product leaks. For now, the plan is geo-restricted to India, but OpenAI has confirmed it is exploring rollouts in other regions.

The timing is strategic: OpenAI revealed that ChatGPT has now surpassed 700 million weekly active users, up from 500 million in March.

India, which CEO Sam Altman recently called the company’s second-largest market, has seen especially strong growth after the rollout of image generation earlier this year.

Data from AppFigures shows that India leads the world in ChatGPT downloads, with over 29 million installs in the last 90 days, though revenues from these users amounted to just $3.6 million in that period.

With its 850+ million internet users, India presents a massive growth opportunity for AI adoption.

Competitors are also targeting this market: Perplexity partnered with Airtel to give away free Pro subscriptions, while Google offered Indian students a free year of its AI Pro plan. OpenAI isn’t giving out freebies, but its localized and affordable pricing may drive stronger subscription adoption in the country.

Towards MCP: Pioneering Secure Collaboration in the Age of AI & Privacy

Towards MCP is a cutting-edge Model Context Protocol platform that lets to connect with any data source and apply intelligence in minutes, and helps you to centrally manage MCP server and client configuration.

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