software verification

Researchers argue that large language models are removing the main bottleneck in formal software verification: the labour of writing mathematical proofs that code behaves correctly. By drafting and repairing proof obligations automatically, LLMs let engineers apply rigorous, machine-checked guarantees to far larger codebases than before. The claim is drawing debate among developers over how reliable such AI-assisted proofs really are.

A developer has released Lasm, an open-source project that lets programmers write formally verified web applications in the Lean theorem prover and run them on Node.js, serverless platforms and edge runtimes. The project, hosted on GitHub, aims to bring mathematical proofs of correctness to ordinary web development, and it is drawing attention among programmers interested in verified software.

A new open-source framework called Witness has been introduced, aimed at verifying how software was built. It helps developers prove the provenance and integrity of their build process, supporting supply-chain security efforts at a time when concerns about tampered dependencies and compromised build pipelines are growing across the software industry.

A developer has presented a new web engine described as both malleable and formally verified, meaning its correctness is backed by mathematical proof. The project invites community feedback on how a provably correct, adaptable browser engine core could change how web software is built and trusted.

Harmonic has introduced Aristotle, a formal verification system designed to mathematically prove the correctness of software written in C and hardware described in Verilog. Formal verification aims to eliminate bugs by exhaustively checking code against specifications rather than relying on testing alone. The release is drawing attention among engineers and researchers interested in applying rigorous mathematical proof techniques to both software and silicon design.

"Vibe coding" — the practice of writing software by describing what you want to an AI assistant and accepting its output with little review — is drawing fresh attention in the US. Related coverage, including a new stealth launch of a code verification tool called Truth, highlights a growing concern: code produced this way still needs rigorous checking before it can be trusted in production.

A developer has launched a specification-first plugin for agentic software development, aimed at making AI coding agents work from explicit specifications and then prove their output against them. The tool, presented on Hacker News, is drawing attention from programmers debating how to keep AI-generated code reliable and verifiable.

The New York Times reports on whether artificial intelligence that has contributed to breakthroughs and new approaches in mathematics could next be used to eliminate computer bugs from software. The piece examines the potential for AI systems to automate code verification and debugging, a longstanding challenge in computer science, and what that shift would mean for how reliable software is built.

Harmonic, the AI startup co-founded by Robinhood CEO Vlad Tenev, is expanding beyond mathematics into software and chip design verification. The company, which pursues formal methods and AI to produce verifiable outputs, will apply its technology to help check the correctness of software and semiconductor designs, a growing need as chips grow more complex.

Zocdoc is allowing AI agents to book medical appointments directly on behalf of patients, extending its online scheduling platform beyond human users. The move signals how healthcare booking is adapting as people increasingly delegate everyday tasks to chatbots and assistants. It raises questions around verification, privacy and accuracy when software handles patient scheduling.

Wired reports that formal software verification — described as 'pain in the ass' math — could be the answer to problems with AI-assisted 'vibe coding', where developers rely on AI-generated code they don't fully understand. The argument: rigorous mathematical methods for proving software correctness could catch the errors and safety risks that come from code written by AI models rather than carefully reviewed by humans.

Apple's iOS 27.2 update introduces the ability for users to confirm their age using a selfie, expanding the company's age assurance tools on iPhone. The feature is drawing attention as tech platforms face growing pressure to verify users' ages and restrict inappropriate content for minors, while privacy advocates watch how the biometric data is handled.

Developer Paul Stack argues that tests written by AI coding agents should be treated as a specification of intent rather than as genuine verification of software behaviour. The post cautions developers against mistaking machine-generated tests for real quality assurance, arguing they only express what the agent assumes the code should do. Software engineers are sharing and debating the piece, weighing how much to trust AI-written tests in their own workflows.

Evergreen is a systems programming language designed with provable safety guarantees, drawing discussion on Hacker News where developers are debating its approach to memory safety and formal verification. Conversation centers on how it compares to Rust and other memory-safe languages, and whether provable safety can be practical for real-world systems work.

Anthropic's Claude AI has been used to generate 180 million lines of formal proofs verifying 123 open source software libraries, according to a report by DevOps.com. The effort represents one of the largest applications of AI-assisted formal verification to date, aiming to prove the correctness of widely used open source code. Developers are watching closely as AI-generated proofs could reshape how software supply chain security is assured.

Debian has released APT 3.3.4, an update to its widely used package manager. The new version introduces SHA3 hashing support, strengthens download verification to protect against tampered packages, and brings improvements to command-line usability. The release is drawing attention among Linux and open-source users interested in software supply-chain security.

Researchers have reconstructed UNK_DeadDrop, a campaign targeting software developers with fake job interviews. The scheme uses ClickFix-style verification pages that secretly swap a malicious command into the victim's clipboard, take-home coding repositories that execute code when opened, and a malicious VS Code extension. The detailed reconstruction of the attack chain has drawn attention within the cybersecurity community.

Harmonic, the artificial intelligence startup co-founded by Robinhood CEO Vlad Tenev, is expanding beyond mathematics into software and chip design verification. The move, reported by Reuters, signals the company's ambition to apply its AI systems to commercially valuable engineering problems, particularly semiconductor design, as demand for AI-driven tools in the chip industry accelerates.

A software developer's remark that software development itself may be pointless has sparked discussion about reproducibility. The argument: open source projects like LibreOffice and darktable produce reproducible builds, meaning anyone can verify the binary matches the source code. With AI-generated code, that guarantee becomes much harder, raising questions about trust and verification in software supply chains.

Australian private equity firm Five V Capital has acquired a majority stake in First AML, an anti-money-laundering compliance startup. The deal delivers a return to early backer Blackbird, one of Australia's prominent venture capital firms, which is exiting part of its holding. First AML helps businesses meet identity verification and AML obligations, and the investment signals continued appetite for compliance-focused software companies.

Anthropic has broadened its Cyber Verification Program's Defense Access tier, making open-source maintainers, nonprofits, and small critical infrastructure operators eligible for the scheme. The move is framed as a response to the reality that much of the software supply chain depends on volunteer maintainers who lack security budgets or resources, prompting discussion about whether AI firms should help underfunded groups defend critical digital infrastructure.

World, the identity project co-founded by Sam Altman, has launched an ID verification service designed for AI agents. The system is intended to confirm that software agents acting online are tied to verified humans or legitimate operators, as automated programs increasingly handle transactions and services. The announcement is drawing attention within the tech and AI community, with debate expected over privacy implications and how digital identity should work as AI agents proliferate.