Committee

Google DeepMind is funding research into the potential dangers of situations where millions of different AI agents interact with each other online. According to Rohin Shah, who directs the company’s AGI safety and alignment research, the mass-market arrival of agents that can carry out tasks without human oversight and follow instructions given to them by other agents creates a whole new class of risk. In an effort to address this, Google DeepMind—which made agent-based tools a centerpiece of Google I/O last month—has teamed up with several other organizations to announce a $10 million funding pot for researchers to study the behavior of multi-agent systems and come up with ways to prevent unsafe scenarios. Joining Google DeepMind are Schmidt Sciences, a philanthropic foundation set up by Eric and Wendy Schmidt; ARIA, the UK government’s moonshot agency; the Cooperative AI foundation, a UK-based nonprofit research outfit; and Google’s charitable arm, Google.org. I asked Shah and James Fox, who leads the Science of Trustworthy AI program at Schmidt Sciences, what they hope to achieve with that $10 million. It’s no small sum, but it’s dwarfed by the budgets commanded by Google DeepMind’s own research teams. The aim is to kick-start research outside tech companies, says Shah: “The strength of academia is that it can look really quite far into the future and do the kind of work that isn’t top of mind at industry labs.” “The main issue is that there just isn’t really a field of research for multi-agent safety yet,” he adds. “And we would like there to be.” The concern is that as more and more AI agents get deployed and begin working together, we could hit a tipping point where imagined scenarios become real. “We see this with humanity, too,” says Shah. “Our institutions can accomplish things that no individual human can.” Shah thinks we have a few more months to go before agents are deployed throughout the economy in numbers that make potential risks a real concern. He wants to get ahead of that moment. Risky business What risks are we talking about, exactly? The possibilities that Shah and Fox have in mind mostly boil down to supercharged versions of bad things that happen on the internet already: scams, prompt injections (where an AI agent is fed malicious instructions, turning it into a self-guiding piece of malware), other forms of cyberattack. We look at what humans do now and ask what the agent version of that would be, says Shah.   “We’ve got this digital commons that is integral to how society works, and you really want to ensure that this doesn’t descend into just absolute anarchy,” says Fox. (I asked Shah if they were considering any worst-case scenarios more on the doomer end of the spectrum, such as widespread economic collapse. “Certainly not if we’re talking by the end of the year,” he said. That’s only six months away! He laughed. “Okay, a while after that.”) Shah and Fox both think that the only way to understand what might happen when large numbers of multi-agent systems interact with each other is to run realistic simulations. They want researchers to drop AI agents into sandboxes and study what they do. You can’t predict what’s going to happen by studying single agents, or even small groups of agents, in isolation. You can’t assume that AI agents underpinned by LLMs will always act rationally, says Fox. And the complexity comes from having huge numbers of interactions at once. Some researchers, including a team at Google DeepMind, have argued that artificial general intelligence (if possible at all) could come not from a single super-smart model but from a kind of agent hive mind, where the capabilities of the whole add up to more than the sum of its parts.   Lack of trust Google DeepMind is not the only top AI firm warning about the risks of the technology it is building. A couple of weeks ago, Anthropic published guidelines for deploying AI agents based on an approach to cybersecurity known as zero trust, which starts with the assumption that a computer system is vulnerable, an agent is an attacker, and a breach will happen. Refael Angel, cofounder and CTO of Akeyless, a cybersecurity firm based in Tel Aviv, agrees that understanding the new risks introduced by agent-based systems is crucial.   Every approach to security in the past has assumed that the machine in question was software written by a human, doing fixed things on fixed paths, says Angel: “An agent breaks all of those assumptions. It reasons, it improvises, and it can be hijacked by a single sentence buried in a document it was asked to read.” Angel welcomes this new funding. “No single lab should author the safety standards everyone else has to trust,” he says. But he cautions that safety researchers can overlook boring problems that are already here in favor of more exotic hypothetical ones. And yet, Fox notes, risks that were hypothetical a few years ago are now very real: “The future’s come more quickly than perhaps expected.”

Anthropic released two models on June 9, 2026: Claude Fable 5 and Claude Mythos 5. Both belong to a tier called “Mythos-class.” This tier sits above the Opus class in capability. Fable 5 is the version claimed to be made safe for general use. Mythos 5 is the same model with some safeguards lifted, kept in limited release. Claude Fable 5 and Mythos 5 Mythos-class models are a tier of Claude models. They sit above the Opus class in capability. The first was Claude Mythos Preview, released in April through Project Glasswing. Fable 5 and Mythos 5 share the same underlying model. The difference is the safeguards. Fable 5 ships with safety classifiers for general use. Mythos 5 has some classifiers removed and stays in limited release. The names reflect this split. “Fable” comes from the Latin fabula, “that which is told.” This is akin to the Greek mythos. The safeguards distinguish the two models, so they carry different names. Anthropic team calls Fable 5 its most capable widely released model. It targets demanding reasoning and long-horizon agentic work. Anthropic states Fable 5’s capabilities exceed any model it has made generally available. Both models support a 1M token context window by default. They allow up to 128k output tokens per request. Pricing is $10 per million input tokens and $50 per million output tokens. That is less than half the price of Claude Mythos Preview. The Capability Case Anthropic reports Fable 5 is state-of-the-art on nearly all tested capability benchmarks. It shows strong results across software engineering, knowledge work, vision, and scientific research. The longer and more complex the task, the larger its lead over Anthropic’s other models. On software engineering, Stripe tested Fable 5 during early access. The model performed a codebase-wide migration in a 50-million-line Ruby codebase. According to Stripe: this took one day. By hand, a team would have needed over two months. Fable 5 is also more token-efficient than past Claude models. On Cognition’s FrontierCode evaluation, Fable 5 scores highest among frontier models. This holds even at medium effort. The eval tests difficult coding tasks under production-codebase standards. On knowledge work, Anthropic cites Hebbia’s Finance Benchmark for senior-level reasoning. Fable 5 posts the highest score of any model there. Gains come in document-based reasoning, chart and table interpretation, and problem solving. On vision, Anthropic calls Fable 5 the new state-of-the-art. It can extract precise numbers from detailed scientific figures. It can rebuild a web app’s source code from screenshots alone. It also needs less scaffolding than prior models. Fable 5 beat Pokémon FireRed with a minimal, vision-only harness. On memory and long-context, Fable 5 stays focused across millions of tokens. It improves its outputs using its own notes. In the game Slay the Spire, persistent file-based memory helped it three times more than Opus 4.8. Mythos 5 carries the science claims. Internal protein design experts accelerated parts of drug design by around ten times. Anthropic also says Mythos 5 is its first model to consistently produce novel scientific hypotheses. Scientists preferred its molecular biology hypotheses around 80% of the time in blinded comparisons. Mythos 5 also ran novel genomics research over a week of largely autonomous work. It trained a custom model on single-cell data spanning 138 animal species. Anthropic says that model outperformed a recent model published in Science, despite being 100 times smaller. How the Safeguards Work Releasing a model this capable carries risk. Without safeguards, Fable 5’s cybersecurity capabilities could be misused to cause serious damage. Anthropic therefore launched Fable 5 with a new set of classifiers. Classifiers are separate AI systems. They detect potential misuse, including jailbreak attempts. They prevent the main model from responding to flagged requests. When Fable 5’s classifiers flag a request, the response is handled by Claude Opus 4.8 instead. The covered areas are cybersecurity, biology and chemistry, and distillation. Users are informed whenever a fallback occurs. For biology and chemistry, Fable 5 falls back to Opus 4.8 on most requests for now. Anthropic cites concern that the same dual-use queries could give uplift to malicious actors. It plans a trusted access program for biology, giving approved researchers Fable 5 without those safeguards. Anthropic tuned these safeguards conservatively. They will sometimes catch harmless requests. On average, they trigger in less than 5% of sessions. Anthropic says more than 95% of Fable sessions involve no fallback at all. For those sessions, Fable 5’s performance effectively matches Mythos 5. Anthropic red-teamed the classifiers extensively. An external bug bounty produced no universal jailbreaks in over 1,000 hours. A universal jailbreak lets a user interact with the model as if its safeguards were absent. Anthropic notes the UK AISI made progress toward one in a brief testing window. Mythos 5 is the same model with cyber safeguards lifted. Anthropic describes it as having the strongest cybersecurity capabilities of any current model. It is deployed through Project Glasswing in collaboration with the US government. Use Cases These capabilities map to several concrete workflows for technical teams: Large-scale code migration: Long-horizon coding suits big refactors and cross-repo migrations. The Stripe example shows this at a 50-million-line scale. Agentic coding pipelines: Fewer turns and token efficiency help multi-step agent runs. GitHub reported autonomy and reliability on complex, long-horizon coding tasks. Finance and analytics work: Strong document and chart reasoning suits senior-level financial analysis. Hebbia and IMC cited gains on reasoning and trading-analysis tasks. Vision-to-code tasks: Rebuilding source from screenshots suits front-end reconstruction and figure extraction. The vision-only harness reduces tooling overhead. Long-running research agents: Persistent memory across millions of tokens suits multi-day research loops. Mythos 5 ran novel genomics work over a week of largely autonomous work. Comparison Table: Fable 5 vs. Mythos 5 vs. Opus 4.8 Attribute Claude Fable 5 Claude Mythos 5 Claude Opus 4.8 Model tier Mythos-class Mythos-class Opus class Underlying model Same as Mythos 5 Same as Fable 5 Opus 4.8 Availability Generally available Limited (Project Glasswing) Generally available Safety classifiers Active (cyber, bio/chem, distillation) Cyber safeguards lifted Opus-level safeguards

Software development has changed. Engineers no longer type most code by hand. They describe intent, and AI agents do the work. Modern tools plan tasks, edit across files, run tests, and open pull requests. Many now ship to production with limited supervision. No single tool fits every need. This guide covers the AI coding agents and platforms shaping development in 2026. Developer Tools Guide Top AI Coding Agents & Platforms — 2026 A practitioner’s field guide to the tools reshaping how software gets built. Development has shifted from typing code by hand to describing intent and letting agents do the work. Today’s tools plan tasks, edit across files, run tests, open pull requests, and ship to production — with limited supervision. No single tool fits every need. This guide walks through the platforms shaping AI-assisted development in 2026 — what each does and where it fits. Use the arrows or dots below to explore → ★ Featured Pick Atoms Goes well beyond a single coding agent. Atoms deploys a coordinated team of AI agents — product management, system architecture, full-stack engineering, SEO, data analysis, and paid advertising. Describe a product in plain language and get a working, deployable app with user logins, data storage, and payments. Race Mode runs prompts across multiple models at once for the best output. 10% off with code MARKTECHPOST10 Try Atoms → Autonomous Engineer Devin AI by Cognition An autonomous AI software engineer, not an in-editor assistant. Give it a natural-language task or a linked ticket; it plans, then executes inside a sandboxed cloud environment with shell, browser, and editor. It runs subtasks in parallel, coordinates sub-agents, and opens pull requests. Best for well-defined bug fixes, features, and migrations. Visit Devin → In-Editor Assistant GitHub Copilot Real-time code suggestions and autocompletion, integrated directly into your editor. It predicts and generates snippets as you type, cutting boilerplate and keeping you in flow — and now extends into chat, pull request summaries, and agentic tasks. A strong default for incremental help inside an existing workflow. Visit GitHub Copilot → UI Building Magic Patterns Builds user-interface components faster from prompts and references. A library of reusable patterns cuts the time spent on repetitive front-end work, so teams move from idea to a working interface prototype with less manual effort and stay focused on the harder parts of a project. Visit Magic Patterns → Agentic IDE Windsurf by Cognition An agentic AI code editor built on a VS Code base. Its Cascade agent reads the whole repository, plans and applies multi-file edits, runs terminal commands, and verifies changes against tests — working across the project as a connected whole. Recent releases added parallel agent sessions and tighter integration with Cognition’s Devin. Visit Windsurf → Prototyping Uizard AI Focused on rapid prototyping for UI/UX designers. Turn text prompts, sketches, or screenshots into interactive prototypes, accelerating iteration and user testing. By lowering the barrier to clickable mockups, it helps teams validate concepts earlier and arrive at more user-centered designs before engineering begins. Visit Uizard → Cloud IDE Replit Agent Brings coding automation into Replit’s browser-based environment. It scaffolds projects, writes and edits code, installs dependencies, and runs apps with no local setup — removing the overhead of configuring a dev environment. Well suited to SME workflows and going from prompt to running app in one place. Visit Replit Agent → Evaluation & Observability Galileo AI An AI evaluation and observability platform rather than a code generator. Its Agentic Evaluations trace agents step by step, score tool-selection quality, detect errors in individual tool calls, and track session success, cost, and latency. Essential guardrails for teams shipping agents to production. Visit Galileo → Terminal-Native Warp An agentic development environment born out of the terminal. Use Warp’s built-in coding agent or bring your own CLI agent (Claude Code, Codex, Gemini CLI). It runs and manages multiple agents in parallel, indexes Git codebases for context, and spans setup through shipping. Available on macOS, Windows, and Linux. Visit Warp → Design-to-Code Lovable Dev Specializes in converting designs into functional applications, bridging design and engineering. By turning visual layouts into working front-end code, it streamlines UI/UX development and tightens the handoff — letting designers and developers collaborate closely and bring designs to life with minimal manual coding. Visit Lovable → Rapid Build Bolt New Known for a friendly interface and easy deployment, accessible to newcomers and experienced developers alike. It supports rapid prototyping in the browser and integrates with common environments, making it a low-friction path from idea to a shareable, running application during fast iteration cycles. Visit Bolt → Multi-Framework UI V0 Dev Supports multiple front-end frameworks, giving developers flexibility to pick the right tools per project. Generated components and interfaces slot into your existing stack, reducing the cost of moving from prompt to usable UI — a versatile choice for diverse applications without framework lock-in. Visit V0 → AI-First Editor Cursor An AI-first editor designed to keep you in control of the codebase. It offers multi-file editing and codebase awareness alongside version control, code reviews, and collaboration — keeping changes organized and maintainable. Built for developers who want substantial AI help while retaining oversight of structure and quality. Visit Cursor → Key Takeaways What to remember AI coding tools have moved past autocomplete — they plan, edit across files, test, and ship. No single tool fits every job — pick by task: autonomous engineer, agentic IDE, evaluation, or full product platform. Atoms stands out for end-to-end product building — a coordinated agent team, not just a code assistant. Atoms ships deployable apps from a prompt — logins, storage, payments, plus Race Mode across models. Recommendation: for the whole product lifecycle, start with Atoms — code MARKTECHPOST10 for 10% off. Try Atoms → ‹ 1 / 15 › MARKTECHPOST Practitioner-first AI/ML news, model releases & developer tools — trusted by 1M+ readers. Atoms* Atoms goes well beyond a single coding agent. It deploys a coordinated team of AI agents. These cover product management, system architecture,

The outspoken longevity scientist David Sinclair has been predicting that one day, you’ll go to the doctor and get a prescription that will make you 10 years younger. Now MIT Technology Review has learned that he has plans to launch human tests of an oral “reprogramming” drug as part of a $101 million competition organized by the XPrize Foundation.  The foundation is offering cash awards to teams able to “restore” a person to an earlier apparent age, as measured by improvements in immune, cognitive, and muscle function.  The grand prize goes to any team able to show a 10-year (or greater) relative improvement after one year of treatment.  Reached by phone, Sinclair, a biologist at Harvard Medical School, confirmed that he plans to give an oral drug mixture to volunteers in a bid to seek “evidence for age restoration in humans.” The trial, if it goes forward, will be a significant new development in the race to harness so-called epigenetic reprogramming. That technology is based on the discovery, 20 years ago, of powerful genes able to turn an adult cell into a stem cell similar to those found in embryos. The age-reversal effect is believed to occur via a resetting of molecular controls on DNA known as epigenetic marks, which help determine a cell’s overall metabolism and identity. Companies are now racing to use that phenomenon for a new form of rejuvenation medicine. Only this January, one of Sinclair’s companies, Life Biosciences, made news by winning approval to launch an initial human trial using a set of powerful reprogramming genes. The company announced today it had treated its first patient.  But that test involves a complex gene therapy and is limited to patients’ eyes, where it could treat conditions like glaucoma.  Sinclair’s new plan is bolder: a reprogramming drug you’d swallow in order to promote such effects across the body.  “What we’re aiming to do is to epigenetically restore the animal and eventually the person,” he says. “It is true that we’ve been doing extensive animal studies with the oral agent and are looking to compete in the XPrize.” This alternative method, chemical reprogramming, uses drugs to mimic the effects of the embryonic genes. That is significant because drug compounds can travel through the bloodstream, reaching most or all cells in a person’s body.  Some experts expressed caution, saying the chemical process, at least as used in labs, is extremely harsh and not even particularly effective. “Who doesn’t dream of whole-body rejuvenation? I think it’s a great goal,” says Sergiy Velychko, founder of Soxogen, a stealth reprogramming company in Boston. “But these chemicals are used in very, very high concentrations for cell reprogramming.” Sinclair declined to describe the exact makeup of the drug candidate, code-named SL-100, calling its contents “highly, highly confidential.” However, he has previously published lab studies of what he called “epigenetic age-reversal cocktails,” which mixed powerful chemicals with known supplements and commercially available medicines.  It’s those latter components that would be easiest to test on people, since doctors are free to prescribe them, even for unusual objectives like age reversal. James Clement, head of Betterhumans, an organization that specializes in life-extension studies using existing drugs, said in a message that he is “running clinical trials” of an oral reprogramming cocktail for Sinclair’s XPrize team. Sinclair’s team is competing in the XPrize Healthspan Competition, launched in 2023. It follows several previous competitions that focused on commercial spaceflight, lunar landings, and other goals. The XPrize Foundation is led by executive chairman Peter Diamandis, also an active promoter of longevity research. “If two teams are equivalent, they would split the award,” says Jamie Justice, a doctor and executive director for the contest, which was bankrolled by Saudi Arabia’s Hevolution Foundation, “But it will be incredibly hard to even get to one winner.” Justice says a judging panel is now in the process of picking 10 finalists from 65 teams that have been exploring health foods, lifestyle interventions, digital trackers, and drug compounds.  Sinclair’s team, Justice says, was a late entrant to the contest, but like all teams, it would be required to move into wider human tests starting this year. “You have to be ready and in trials,” she says. The race to harness the reprogramming phenomenon and apply it to living people is heating up, even outside the XPrize competition. On June 2, a startup called NewLimit, founded by the crypto billionaire Brian Armstrong, said it had raised a further $435 million, from investors including Peter Thiel’s Founders Fund, to support what it calls “age reprogramming.”  The company says it is working toward delivering genetic reprogramming instructions to the liver, to treat diseases of that organ. But Sinclair has been saying that whole-body rejuvenation is a possibility too. And for that, chemicals, rather than gene therapy, could be the most practical strategy.  Sinclair says his lab has been searching for such compounds and is starting to use AI “to improve the oral agents that we’re testing.” Chemical reprogramming cocktails, as used in labs, typically involve a mix of vitamins, approved drugs, and experimental molecules. For instance, one recipe Sinclair filed a patent on includes the supplement forskolin,  the antidepressant tranylcypromine, and an experimental chemical, laduviglusib, which has been tested against Alzheimer’s, among other ingredients. “In those days it was a six-factor cocktail,” Sinclair says of his earlier research. “But we’ve come a long way. I can’t disclose what’s in it, but it’s an improvement and an advance on that, and we’ve done a number of animal studies. They are not published, but we’ve been doing them for a long time, and we want to make sure that we’ve done a full investigation of safety and efficacy before we release any of the data.” While Sinclair’s results aren’t published, other teams say attempts to reverse the age of entire animals using chemical drugs haven’t worked yet. Last year, the lab of Vadim Gladyshev, another Harvard biologist and a member of a different XPrize team, reported on its attempt to