Author: IBL News

IBL News | New York

Anthropic revealed that 80%+ of the code merged into its codebase as of May 2026 is authored by Claude. 3 (in February 2025, this number was in the low single digits), while it detailed its progress toward recursive self-improvement and its implications.

“At Anthropic, we are delegating a growing share of AI development to AI systems themselves, which is speeding up our work.”

The company believes that the so-called recursive self-improvement practice – with AI systems capable of fully autonomously designing and developing their own successors— could arrive sooner than most institutions are prepared for.

Today, autonomous agents can now run code themselves and delegate hours of work to other agents.

Systems capable of fully building their own successors would represent a major development in science, healthcare, and beyond, but might increase the risk of humans losing control over AI systems.

In the future, agents could become capable of building and training models themselves. If this happens, future versions of Claude could be continuously improved by Claude itself.

The Anthropic Institute is showing that AI is already accelerating the development of AI systems. To take just one example: today, Anthropic engineers on average ship 8x as much code per quarter as they did from 2021 to 2025.

The technical trends discussed in this piece suggest that AI systems are going to become much more capable in the coming years. These trends have huge implications.

The evidence points out that the rate at which AI models improve is accelerating, with the length of tasks doubling roughly every four months, up from an earlier trend of doubling every seven months.

In March 2024, Claude Opus 3 could complete software tasks that would take humans about 4 minutes. A year later, Claude Sonnet 3.7 managed tasks that would have taken about an hour and a half. A year after that, Claude Opus 4.6 managed 12-hour tasks.1 If this trend holds, tasks that take a skilled person days could come within range this year. In 2027, AI systems could be capable of tasks that take a person weeks.

Our internal data shows Claude is accelerating AI development—a possible path to recursive self-improvement, or AI autonomously building a more capable successor.

It’s happening faster than we thought, and the implications deserve greater attention. https://t.co/OVVPJO7VQx

— Anthropic (@AnthropicAI) June 4, 2026

IBL News | New York

The Chinese model MiniMax M3, natively multimodal and with a context window of up to 1M tokens, was released this month. It’s an open-weight model, meaning its trained parameters are publicly available for download, local deployment, and fine-tuning.
However, the company withheld the original training code, training data pipelines, and specific inference operators. It can run on a desktop computer.

Users can use the model via MiniMax M3, MiniMax Code, the Token Plan, and API services.

According to the company,

• “On SWE-Bench Pro, which measures coding capability, MiniMax M3 surpasses GPT-5.5 and Gemini 3.1 Pro and approaches Opus 4.7. On SVG-Bench, a benchmark that comprehensively evaluates SVG generation performance, MiniMax M3 surpasses Opus 4.7.”“On OmniDocBench, a multimodal benchmark, MiniMax M3 scores above Gemini 3.1 Pro. On Claw-Eval, an end-to-end evaluation framework for autonomous agents, MiniMax M3 achieves the highest score.”

• “For long-horizon complex tasks, MiniMax Code’s Agent Team can break large tasks down into multi-stage, concurrent, and dynamically adjustable workflows, which are then advanced collaboratively by a cluster of agents. Through a Producer + Verifier adversarial harness loop, the Agent Team can continuously produce, reflect, and correct itself during execution. It can run autonomously for days without human intervention and ultimately deliver high-quality results.”

• “We have seen that Claude Code has also recently released Dynamic Workflows in a similar direction. Compared with Claude Code’s stronger emphasis on fixed orchestration based on JS code, MiniMax Code focuses more on “deep reflection and continuous error correction”: the agent adjusts its plans and priorities in real time based on task progress, while users can step in at any time to add requirements or correct the direction.”

On the other hand, the company announced that MiniMax Code, “built on a harness based on the outstanding open-source community projects OpenCode and Pi”, will be open-sourced this project in the future.

IBL News | New York

Paris-based AI company Mistral AI is positioning itself as a critical European counterweight to U.S. dominance in the race for superintelligence or AGI. The start-up is growing by playing to its home crowd: selling services hosted in European data centers, independent of American and Chinese competitors, to local governments and enterprises.

Experts say that U.S. tariffs and President Trump’s threats to take over Greenland have given new impetus to a European push for tech independence.

Guillaume Lample, Mistral’s co-founder and chief scientist, said, “Very soon in the future, we are probably going to see AGI or superintelligence, so it is very important that we have access to these models also in Europe.”

“Also, as long as we have adversaries that are threatening, we do need to have our own capabilities,” he said. “Europe in particular needs to have strategic autonomy when it comes to defense systems.”

Founded by three French AI researchers from Google and Meta Platforms, Mistral was valued at around $14 billion last year and promises more than $1 billion in revenue this year, a fraction of the size of its Silicon Valley rivals.

Mistral has already landed deals to supply AI services to European corporate heavyweights such as Airbus and BMW, as well as the French military. Mistral also said it was building a new 10-megawatt data center south of Paris, part of the $4.7 billion it is plowing into data centers in France and Sweden.

However, the company acknowledges that a big obstacle to tech independence is the scale of investment necessary.

“We can’t put 50 billion [dollars] on the table to build a gigawatt ahead of demand,” the executive said. “That’s potentially our biggest bottleneck.”

IBL News | New York

OpenAI confidentially filed for a U.S. IPO (Initial Public Offering) yesterday, targeting a valuation of up to $1 trillion.

OpenAI, the creator of ChatGPT, is set for a stock market debut, which could come in September, intensifying competition for investor capital, especially as rival Anthropic aims to go public this year as well.

OpenAI did not disclose the size or terms of the offering, and said a timeline has ‌not yet been determined.

“It may be a while because there are things we want to do that are likely easier as a private company,” it said in a statement.

At the mentioned valuation, OpenAI would be the third trillion-dollar-valuation company (along with SpaceX and Anthropic), testing investor appetite in the most consequential quest for high-growth technology stocks in ​the last 10 years.

On June 1. Anthropic, the company behind Claude Code, confidentially filed for a U.S. initial public offering, weeks ⁠after raising $65 billion in a funding round that valued it at $965 billion.

IBL News | New York

As it races to catch up with rivals, Apple unveiled its long-delayed AI-powered version of Siri, its digital assistant, at its annual Worldwide Developers Conference (WWDC) in Cupertino, California, on Monday.

The new Siri, called Siri AI, includes a variety of upgrades that bring it closer to AI helpers like OpenAI’s ChatGPT, Google’s Gemini, and Anthropic’s Claude, enabling users to have back-and-forth conversations and complete tasks. It’s the biggest update to Apple’s Siri in years, turning it into an AI hub for iPhone, iPad, and Mac.

Siri AI is powered by Google’s own Gemini AI model. This is also the first time that the company has officially confirmed that some of its Apple Intelligence features will run on Nvidia chips.

This redesigned Siri can speak back and forth with the user, a major improvement over previous versions of the assistant. It has its own dedicated app and customizable voice options.

In a demo at WWDC, Siri was able to check concert dates, set a reminder to buy tickets, and even get directions to pick up a friend on the way to the concert venue.

Apple unveils Siri AI by Apple Intelligence. pic.twitter.com/VClr2s8RrA

— Yahoo Finance (@YahooFinance) June 8, 2026

IBL News | New York

Ultra billionaire Jeff Bezos dismissed fears of artificial intelligence-fueled job displacement, arguing that the technology will augment workers, boost productivity, and lead to labor shortages and deflation across a range of goods and services — but only if “we let this technology play out and don’t hamstring it with regulation too early.”

His optimism contradicts the negative public perception of AI, which has worsened significantly in recent months.

A recent Pew Research Center found that half of U.S. adults are more concerned than excited about the increased use of AI in daily life.

Respondents pointed to the potential harms of AI on creativity and relationships and expressed pessimism about its impact on education and jobs.

Rapid development of AI data centers, which can span hundreds of thousands of square feet, has also sparked widespread backlash over their impacts on nearby residents’ lives.

On the fear of increased mass layoffs. of software engineers and programmers due to the rise of AI coding tools such as Claude Code and Cursor, Bezos argued that these tools help programmers identify and solve problems in their work.

“It’s just that the work is going to be done at a higher level,” Bezos said. “It’s going to be done with a bulldozer instead of a shovel, and that’s going to be a good thing.”

Bezos believes “President Trump has been right about a lot of things.”

IBL News | New York

NanoClaw creator Gavriel Cohen and his brother and co-founder, Lazer Cohen [in the picture above], whose project was created and launched in under six weeks, declined a $20 million acquisition offer and instead raised an oversubscribed $12 million seed round for NanoCo, a secure alternative to OpenClaw.

The funding was led by Valley Capital Partners and included participation from Docker, Vercel, Monday.com, Slow Ventures, and angels such as Clem Delangue, CEO of Hugging Face and creator of the open-source tiny robot Reachy Mini.

Gavriel Cohen said that about 50 or more founders and tech executives sent DMs asking to invest.

Interest in NanoClaw went viral after AI researcher Andrej Karpathy tweeted his praise for it. The project skyrocketed after Singapore’s foreign minister called NanoClaw his “second brain” in a Facebook post.

Instead of running directly on a computer, with access to all services and credentials, NanoClaw runs sandboxed in a container — a practice that is becoming a common solution to running more secure, OpenClaw-like setups.

NanoClaw has many thousands of users. Early adopters include executives with technical skills at tech companies, such as Amazon, Gap, Google, Meta, SentinelOne, and Accenture

NanoCo is offering implementation services, these days known as “forward-deployed engineers,” to help businesses roll out NanoClaw AI agents to employees and provide ongoing support.

IBL News | New York

Given that a single agent cannot handle large tasks, such as a complex migration, security audit, or bug hunt, especially in legacy codebases, Anthropic introduced “Dynamic Workflows” in Claude Code last week, enabling orchestration scripts that run tens to hundreds of parallel subagents in a single session.

Dynamic workflows, which consume substantially more tokens than a typical Claude Code session, are available in research preview in the Claude Code CLI, Desktop, and the VS code extension for Max, Team, and Enterprise (if admin enabled) plans, as well as on the Claude API, Amazon Bedrock, Vertex AI, and Microsoft Foundry.

The company provided this explanation: “Dynamic workflows are built for parallel and long-running work that can extend into hours and days, doing the most complex engineering work that previously would have taken weeks. Progress is saved as the run goes, so a job that’s interrupted picks up where it left off instead of starting over. Because the coordination happens outside the conversation, the plan stays on track no matter how big the task gets.”

On a Max or Team plan, or using Claude Code via the API, dynamic workflows are on by default.

Another May launch mentioned the possibility of Claude Managed Agents operating in self-hosted sandboxes and connecting to private Model Context Protocol (MCP) servers.

These sandboxes can run on owned infrastructure or with managed providers like Cloudflare, Daytona, Modal, or Vercel.

Self-hosted sandboxes (for keeping sensitive files, packages, and services) are available in public beta on the Claude Platform, and MCP tunnels are in research preview.