Anthropic has released the latest version of its mid-sized model, Sonnet 5, which the company claims is its most “agentic” yet. For developers writing agents to automate tedious and recurring tasks, Sonnet 5 promises improved capabilities in reasoning, tool use, coding, and knowledge work. This version is also less likely to pull embarrassing (for Anthropic) gaffes of misunderstanding, so the company asserts. “Our safety assessments found that Sonnet 5 shows an overall lower rate of undesirable behaviors than Sonnet 4.6, and is generally safer to use in agentic contexts,” the company asserted in an introductory blog post on Tuesday. Sonnet 5 is smarter at refusing malicious requests and resisting prompt-injection attempts. It doesn’t hallucinate as often and doesn’t suck up to the user so much (“sycophancy”) as did its older brown-nosing Sonnet 4.6 sibling. It is also more aware of, and can block, user misuse and deception, the benchmarks in Anthropic’s System Card seem to indicate. Sonnet is the default model for Claude Free and Pro users, and is also available to the token-pinching Max, Team, and Enterprise customers. The benchmarks also indicate Sonnet 5’s performance can come close to that of Anthropic’s flagship enterprise-focused Opus 4.8, but can execute the same tasks more cost effectively. For Opus, Anthropic charges $5 per million input tokens and $25 per million output tokens. Starting in September, Sonnet users will pay $3 per million input tokens and $15 per million output tokens, though Anthropic is running a special through the end of August where tokens will only be $2 per million inputs and $10 per million outputs. So users trimming their token budgets can run jobs through Sonnet instead of Opus, the company suggests. The 5.0 release offers a new setting to adjust the model’s effort at completing tasks. Simple tasks can be completed through one of the lower “effort” settings, which uses fewer tokens, while longer-running agent-based tasks can go full throttle (“xhigh” or even Homer Simpson’s favorite setting, “max”). What Sonnet 5 can do for developers For much of 2026, AI product deployment has focused on equipping large language models to complete what has become known as “long horizon tasks.” It might be easy for a model to fix a bug or churn out some code. However, keeping its finicky attention fixed on a multi-part task has proven more difficult. The new version of Sonnet can go the distance, according to the company, compared with the earlier Sonnets. “Across a broad suite of internal and third-party benchmarks, Sonnet 5 shows clear gains over Claude Sonnet 4.6 in coding, agentic search, multimodal reasoning, and professional-task performance,” the System Card asserted. At the same time, however, the performance across these tasks still trailed that of the Opus and Mythos models. One testimonial from a Zapier engineer described a two-part job that flummoxed earlier Sonnets: Update a contact database and send out a notice to all users. Version 5 was able to complete the task “end to end.” Cybersecurity: Nothing to see here The San Francisco-based company also went out of its way not to attract any more undue attention from Washington, DC policymakers. “We did not deliberately train Sonnet 5 on cybersecurity tasks,” the company asserted. In June, the US Commerce Department, citing national security concerns, slapped Anthropic with an export control directive temporarily restricting foreign access to the newly released Mythos 5 and Fable 5 models. Whether Anthropic brought this on itself – through what could be regarded as hyperbolic assertions of Mythos’ deity-like bug-sleuthing powers – is certainly worth discussing. But Anthropic, like Pete Townshend, certainly won’t be fooled again. While it can readily perform routine cybersecurity tasks, Sonnet 5 is guardrailed against generating offensive attack code. When commanded to write a Firefox exploit, it failed to complete the task (though it got a bit further than Sonnet 4.6 in the attempt). “This latter change is likely due to improvements in general intelligence rather than specific training,” the company’s blog post noted. ®

The Homebrew team has released version 6.0 of this popular open-source package manager for macOS and Linux, with a new mechanism for trusting packages and support for sandboxing on Linux, to align with existing sandboxing on macOS. Homebrew 6.0 introduces tap trust, a "tap" being a collection of formulae, casks (a package of pre-compiled binaries) and commands which usually reside in a Git repository. The tool trusts official Homebrew taps by default, but requires an explicit agreement before it will trust third-party taps (which can include arbitrary Ruby code) before they install or run any code. Tap trust is part of Homebrew’s approach to supply chain security, which has a number of distinctive features. Package maintainers are Homebrew maintainers, not the authors of the package. Names are maintainer-curated, so typosquats (giving a package a misleading name designed to be similar to one that is popular) can be rejected. Each download is pinned to a sha256 checksum. Package binaries are built from source, which protected Homebrew from incidents like the Trivy compromise earlier this year when official Trivy binaries were replaced with malicious versions. These and other features of Homebrew security are described in the documentation. Project leader Mike McQuaid told us that "Homebrew was less vulnerable 10-15 years ago than npm is today. The trust model is radically different and, even today, we are much quicker to break backwards compatibility in the interest of security." A new security feature is sandboxing on Linux when Homebrew compiles software. This was already implemented on macOS (and has been for a decade). Version 6.0 adds a Linux implementation based on the Bubblewrap project, and this will be on by default for developers. A new Homebrew sub-command, brew vulns, will check installed packages for known vulnerabilities, by checking against the OSV (vulnerability database for open source). The commands brew install and brew upgrade will now show a dependency summary and require a conformation prompt before running, called ask mode, following a developer survey earlier this year where this was highly requested. Another new command, brew exec, will run a Homebrew-provided executable, similar to the way npx works for npm packages. Homebrew startup performance in 6.0 is said to be faster, thanks to parallelised bottle fetching (a bottle is a pre-built package) and other optimizations. Apple is phasing out support for Intel macOS both for future versions of macOS and for Rosetta, the Intel compatibility layer. Homebrew is following: in September this year no new bottles will be built for macOS Intel and from September 2027 macOS Intel will be "unsupported entirely and all related code deleted," according to the post introducing Homebrew 6.0. Homebrew is well-liked by developers, and becoming more popular on Linux as well as macOS. There is some frustration though regarding the dropping of Intel support. "The deprecation of Intel support is agressive! Every Mac enthusiast I know who uses a Mac as a server uses their old machines, which are pretty much all Intel. We'll lose support from you guys a year before Apple!," said one. McQuaid replied noting that Homebrew will still work for a year after support is dropped to "Tier 3”, meaning almost unsupported, and added that "there’s nothing stopping you for doing the work to setup ‘Intelbrew’ and support it for the community." Another issue he mentioned is that GitHub is dropping macOS Intel runners for continuous integration towards the end of 2027. It is notable that Homebrew 6.0 made extensive use of AI coding. A document on responsible AI usage takes the line that AI contributions must be disclosed and human-reviewed, and that AI is not responsible for any code, rather the human contributor is responsible. "AI is great if used responsibly which means a human reviewing all changes both before PRs submitted and a maintainer reviewing before PRs are merged. I have found despite using it responsibly it has been a huge personal accelerator," McQuaid told us. ®

GitHub will change npm's defaults so the install command no longer runs scripts automatically, disabling a feature commonly exploited by malicious packages such as the notorious Shai-Hulud worm. Maintainer Leo Balter said: "Install-time lifecycle scripts are the single largest code-execution surface in the npm ecosystem. Every npm install runs scripts from every transitive dependency, so a single compromised package anywhere in your tree can execute arbitrary code on a developer machine or CI (continuous integration) runner." In npm 12, due July, three security-focused defaults are changing. Scripts configured for preinstall, install, or postinstall will no longer run unless explicitly permitted via allow-scripts. The --allow-git flag, which pulls dependencies from remote URLs, will default to off, closing an attack path where a malicious .npmrc file could override the Git executable and achieve arbitrary code execution. Finally, allow-remote will default to none, blocking dependency downloads from remote URLs entirely. It will still be possible to allow scripts to run via an allowlist in the package.json configuration file. This will be pinned to the installed version of a package by default. These are breaking changes, and Balter recommended developers run the commands to allow scripts for every currently installed package in a project that requires them. "This gets you protected against new, unexpected scripts immediately," he said. The next step is to review these packages and deny scripts for those where they are not needed. Some packages require script approval to function, including native modules that compile on install, testing tools like Playwright and Puppeteer (which fetch binaries via postinstall), and Electron, which wraps the Chromium browser engine for cross-platform desktop applications. These features have been available since npm version 11.10.0, released in February, but as opt-in flags rather than defaults. That version also introduced min-release-age, which blocks installation of package version newer than a specified number of days, designed as a safeguard against newly published malicious packages. Best security practice for developers using npm 11.16, the current version, is to set these flags on in .npmrc or via environment variables, which will also prepare a project for the changes in version 12. One annoyance is that the existing flag ignore-scripts does not support an allowlist, other than via an additional tool. The ignore-scripts setting will override allow-scripts, so developers will need to remove it, if set to true, to enable approved scripts to run. The allowScripts setting exists in npm 11 but is advisory only. Will this fix npm security issues? Unfortunately not. "Now all the malware can move from the install script to the module itself where it will inevitably still be run," said one developer. Another common view is that developers should use pnpm, which already has safer defaults than npm, including a minimum release age. There is consensus, though, that these changes do improve npm security and are long overdue. The pull request for this change includes the remark that "npm is the only remaining major package manager that runs dependency install scripts by default. pnpm v10+, Yarn Berry, Bun, and Deno all block them." ®

Welcome to GoGatoZ — a purpose-built Go tool for GitLab CI/CD security auditing that can perform and automate the entire CI/CD kill chain along with everything those one-off scripts did and then some.

The post Auditing GitLab: The CI/CD Kill Chain appeared first on Black Hills Information Security, Inc..

The Challenge As stated in PART 1 of this blog, the Windows endpoint defense technology stack in a mature organization represents a challenge for Red Teamer initial access operations. For […]

The post Initial Access Operations Part 2: Offensive DevOps appeared first on Black Hills Information Security, Inc..