RSS Feeds

From state and local food-safety response to the federal health agencies that fund that work, a record-breaking year for cyclosporiasis cases runs up against the Trump administration's cuts last year.

Есть компании, где вопрос «какую модель выбрать» отходит на второй план, а первым встаёт другой: выйдут ли данные за периметр. Банк, госкорпорация, оборонный подрядчик, медицинский холдинг. И если ответ «выйдут, но мы всё обезличим» — начинается тяжёлый разговор. Обезличивание не даёт технической гарантии: достаточно одного не вычищенного поля в выгрузке или сочетания признаков, по которому человек восстанавливается. А отвечать за это решение будет конкретный человек — и чем выше цена ошибки, тем охотнее он выбирает вариант, где данные просто никуда не уходят.

Крупные игроки эту задачу решают. Сбер сделал GigaChat, у Яндекса своя линейка, ряд банков строит внутренние платформы; другие компании разрешают облачные модели, но под контролем — DLP, договоры, фильтрация персональных данных на выходе. Оба пути рабочие. Только первый — это программа на сотни человек и бюджет соответствующего порядка, а второй упирается в политику безопасности, которая не везде это позволяет.

А между «построить свою модель, как Сбер» и «отправлять всё в облако» лежит широкая полоса: компании, которым первое не по карману, а второе не разрешает служба безопасности. Вот им обычно и говорят: тогда ничего серьёзного не получится. Аргумент — открытая модель среднего размера проигрывает облачным гигантам почти по всем строкам бенчмарков, так что максимум, на что можно рассчитывать, — умный автодополнитель.

Вот это утверждение я и взялся проверить на практике: собрать не чат‑бота, а полноценный корпоративный оркестратор — систему, которая понимает задачу, сама выбирает исполнителей, ходит в корпоративные системы, делает многошаговую работу и отвечает файлом, письмом или отчётом. И запустить его не на кластере ускорителей, а на одной потребительской видеокарте.

The attorney for Clancy, who strangled her three children, says she suffered from postpartum psychosis

Hundreds of Lindsay Clancy supporters, many wearing pink clothing emblazoned with phrases such as “Believe”, “She Needed Help” and “Peace For Lindsay”, gathered on Thursday outside the courthouse where she is on trial in the killings of her three children.

Several of the 300 women, and a few men, said Clancy’s story resonated with them and that they wanted to raise awareness about how the mental health system treats women. Clancy’s lawyer does not dispute that she killed the children, but says she should not be held criminally responsible because she was mentally ill.

В данной статье рассматривалось: что такое шейдеры, что такое вычислительный шейдер, как его создать и использовать.

Hundreds of supporters, many wearing pink clothing, gather outside court in Massachusetts as trial continues

Hundreds of Lindsay Clancy supporters, many wearing pink clothing emblazoned with phrases such as “Believe”, “She Needed Help” and “Peace For Lindsay”, gathered on Thursday outside the courthouse where she is on trial in the killings of her three children.

Several of the 300 women, and a few men, said Clancy’s story resonated with them and that they wanted to raise awareness about how the mental health system treats women. Clancy’s lawyer does not dispute that she killed the children, but says she should not be held criminally responsible because she was mentally ill.

• Meetings around event seen as chance to lobby for votes

• Fifa president may meet his most credible challenger

Gianni Infantino wants to split the European and North American coalition that is seeking to oust him by attending a junior tournament in the Dominican Republic this weekend in another gamble to cling on to power from the Fifa president.

The Guardian has learned that Infantino has made plans to attend the U14 Boys Challenge competition organised by the Caribbean Football Union (CFU), 21 of whose associations are voting members of Fifa, despite being urged to stay away from the event by senior figures within Concacaf – the Confederation of North, Central America and Caribbean Association Football – which is working with Uefa to remove him.

New managers galore suggest many of the usual suspects will struggle to dethrone the champions, but their own hunger could be the key

Mikel Arteta wants to get his messaging absolutely right. According to him, Arsenal are not about to defend their Premier League title. They are set to attack another one.

Remember Win the dog, who the manager brought in to the club’s training ground a few years back? She must now be called “Won”. But it is when Arteta quibbles over another bout of semantics that he highlights why he is entering the new season with such energy and optimism.

In some tight midterm races, how farmers frustrated with the war and rising costs vote could make a difference.

(Image credit: Michelle Gustafson for NPR)

Custom Flows already let teams turn manual, multi-step work into automation that runs on GitLab events. But writing one meant learning the Flow Registry schema first. That requirement is a real barrier. The people who understand a workflow best aren’t always the same ones who know a YAML schema.

In GitLab 19.3, the Flow Creator agent removes that requirement. Describe the flow you want in your own words, and it responds with a complete, runnable definition.

When good automations stopped at the schema

Consider two people who know exactly what should be automated, but not how to build it.

• The security analyst triages vulnerability reports every day and knows which patterns are false positives and who owns each affected dependency. None of that knowledge is captured in a schema they have never used.
• The planning lead knows what should happen next when a work item changes status. The automation is obvious to them. The Flow Registry syntax is not.

Writing a flow meant learning that schema before any of this knowledge could become automation. That requirement limited who could author flows to people with schema familiarity, not people with process knowledge. The result is a gap between who understands the work and who can automate it. That gap effectively slows down automations across the entire software lifecycle especially where bottlenecks already lurk.

Bring flow building to everyone who understands the work

Flow Creator is a foundational agent within GitLab Duo Agent Platform, available via Agentic Chat. Invoke the agent, describe what you want the flow to do in plain language, and it produces a complete flow definition.

For example, "When a work item is created with the label security-triage, check it against our known false-positive patterns, find the affected dependency and its owner, and post a severity assessment as a comment." This simple explanation gives the agent a trigger, steps, and an output target.

You don't have to get the description perfect on the first try either. If something is ambiguous or missing, like which project the flow should target or whether a step needs human approval, the agent confirms before it builds instead of guessing and returning a flow that runs against the wrong project or skips a needed approval.

Guardrails hold as the author pool grows

More people building flows does not mean looser guardrails. Every flow runs through a scoped service account under composite identity, so it can never reach further than the permissions of the person who ran it. That account is scoped to the top-level group and cannot access resources outside of its bounds. Authoring a flow and enabling it stay separate: Anyone can describe a flow to Flow Creator, but enabling one still requires the Maintainer role or higher. Group Owners and instance administrators can turn Custom Flows off entirely.

See Flow Creator in action:

How it works

Flow Creator does three things behind the scenes:

• Reads current documentation. The agent checks the Flow Registry documentation before every response, rather than answering from what it learned during training.
• Applies rules. These rules encode failure patterns seen in real flows, so common mistakes get caught before a session even starts.
• Runs a pre-output checklist. This check happens before any YAML is generated, catching issues early rather than after registration.

Each rule corresponds to ways flows break today:

• Missing project_id. The flow runs but has no clear target, so it acts against the wrong project or against nothing.
• Broken human in the loop gates. A flow meant to pause for approval proceeds without it.
• Wrong sends_response_to targets. The flow completes and writes its output somewhere nobody is looking.
• Missing stopping instructions. The agent keeps working beyond task completion.

The output is a full, runnable definition, not a fragment to assemble. Paste it into the configuration editor when you create a flow from a project or from the AI Catalog. Then, choose whether the flow is private, restricted, or public, and add the triggers that should start it.

Run your first custom flow

Flow Creator ships in GitLab 19.3 as part of GitLab Duo Agent Platform. If you have been waiting on someone else to write your automation, you don’t have to anymore.

Read the Custom Flows documentation to see how to register and enable a flow, then describe the one you want to Flow Creator. Start a free trial of GitLab Duo Agent Platform. On the Free tier, you can sign up in a few simple steps. If you're already on GitLab Premium or Ultimate, you can turn on Duo Agent Platform and use the GitLab Credits included with your subscription.

Many enterprises choose GitLab Dedicated for a clear reason: a single-tenant instance, managed by GitLab, in a cloud region they select. That isolation already covers source code, project data, and the software development lifecycle they run on GitLab.

GitLab Dedicated customers can now deploy the AI Gateway for the GitLab Duo Agent Platform inside their single-tenant SaaS infrastructure, so AI-processed data stays in their environment and chosen region.

GitLab Dedicated as the critical infrastructure for software delivery

GitLab Dedicated is deployed in your preferred AWS region, and hosted and maintained by GitLab. You can bring your own encryption keys. No private cloud installation required.

GitLab operates the instance, including high availability and disaster recovery, and maintains 99.9% monthly availability for core services. Updates to GitLab Dedicated follow a stability-first cadence, with instances on the previous minor version (N-1) relative to the current GitLab release.

That is the infrastructure organizations already trust when they scale software delivery under audit: guardrails you can take to a regulator, SaaS operations you do not staff, and an availability target that holds when the platform is in the critical path.

Trust should travel with every agent

Agentic workflows demand more from your software development lifecycle beyond the AI coding: increased pipeline volume, greater risk for vulnerabilities introduced, context that needs to travel with every agent and flow, among others.

The isolation, residency, and service-level requirements that already apply to GitLab Dedicated should apply to that inference path too, especially with growing adoption of agentic software development.

Adopt AI while staying compliant

Here is what's new with the introduction of AI Gateway for GitLab Dedicated:

• Keep AI processing in the tenancy. The AI Gateway runs in the Dedicated single-tenant environment, aligned with data residency.
• Bring your own models. Connect models you control so inference follows your policy, not a one-size default outside your region.
• Unlock Duo Agent Platform where you already run GitLab. Agentic work becomes available under the Dedicated deployment model your auditors already understand.

GitLab Dedicated customers using GitLab Duo Agent Platform can connect the AI Gateway to Amazon Bedrock as their model backend, so inference stays in their AWS region. Or you can use your own preferred model providers.

If you have specific model requirements, data classification rules, or existing commitments, you can identify models hosted in Amazon Bedrock for your team to use with GitLab Duo Agent Platform. All capabilities of Duo Agent Platform, including its top use cases such as automating custom code reviews, fixing failed pipelines, and security analysis can now run against your chosen models.

Start your agentic software delivery on GitLab Dedicated

If you are on GitLab Dedicated, review how to add a self-hosted model, validate model connectivity in a non-production project, and then expand agentic workflows under your existing change controls.

If you are not using GitLab Duo Agent Platform today, you can start with a free trial. If you are an existing subscriber to GitLab Dedicated, you can get started simply by turning on Duo Agent Platform and start using the GitLab Credits that are included with your subscription.

Security teams have historically struggled to keep up with triage and remediation when development was happening at human speed. Today, that challenge is exacerbated by developers writing and shipping code at machine speed with AI.

At the same time, AI is helping attackers find and exploit vulnerabilities faster in production. Verizon's 2026 Data Breach Investigations Report found vulnerability exploitation overtook credential abuse as the top breach entry point. Vulnerability exploitation accounts for 31% of breaches versus just 13% in previous years, and only 26% of known exploited vulnerabilities were remediated this year, down from 38%. That means more breaches are starting with a vulnerability that was already sitting in production, unpatched.

In GitLab 19.3, teams can now run Static Application Security Testing (SAST) False Positive Detection and Agentic SAST Vulnerability Resolution, in bulk, on their existing vulnerability backlog.

Dismiss false positives and remediate vulnerabilities in bulk

Most teams have a known vulnerabilities backlog already running in production, far more than any team has time to open and triage one at a time. That used to be a workload problem. Now it's a timing problem too: The same AI accelerating development is accelerating exploitation, which means findings that sit in the backlog longer carry more risk than they used to.

GitLab addresses this with two layers of automated triage and remediation. The first covers active vulnerabilities in production. With GitLab 19.3, teams can bulk-select multiple vulnerabilities to run agentic flows, SAST False Positive Detection and Agentic SAST Vulnerability Resolution, against the vulnerability backlog in production. SAST False Positive Detection analyzes a finding's surrounding code and returns a confidence score and explanation for whether it's a real risk or a false positive. Agentic SAST Vulnerability Resolution takes findings confirmed as real risk and opens a ready-to-merge fix that remediates the vulnerabilities, so a developer reviews and merges the fixes instead of writing it from scratch.

This applies to all SAST vulnerabilities in the vulnerability report, no matter the scanner origin or severity. You can import findings from any third-party scanner that outputs SARIF; GitLab ingests them into a unified vulnerability report.

With bulk SAST False Positive Detection and Agentic SAST Vulnerability Resolution teams can cut years of accumulated risk in a single action instead of one vulnerability at a time and to reduce the risk of vulnerabilities being exploited in production.

See these two features in action:

The second layer covers any new vulnerabilities identified. Teams can configure pipelines to run SAST False Positive Detection and Agentic SAST Vulnerability Resolution automatically on Critical and High severity vulnerabilities (watch an interactive demo). New findings get triaged automatically, and when one is deemed a real risk, GitLab creates a ready-to-merge fix before anyone has to open the finding by hand.

Start automating triage and clearing your backlog today

With bulk triage and remediation developers spend less time confirming false positives and more time shipping fixes that drive down real risk.

Start a free trial of GitLab Duo Agent Platform to try it today.

FAQ

Do bulk SAST False Positive Detection and Agentic Vulnerability Resolution consume GitLab Duo Agent Platform credits?
Yes. Please see the pricing page for further details.

Does bulk triage and remediation cost more credits than running detection one at a time?
No. Each finding still costs the same per-execution rate; bulk just queues many at once instead of triggering them manually one by one.

Which vulnerabilities qualify for bulk triage?
Any SAST vulnerability in your vulnerability report, no matter the scanner origin or severity.

Will running this on my whole backlog overwhelm my pipeline?
No. Concurrency limits keep the bulk job from competing with your pipeline's other work, so clearing thousands of findings doesn't come at the cost of build capacity.

Can I stop a bulk job once it's running?
Yes. A progress indicator tracks the job while it runs, and you can cancel it at any point.

DHS says Manuel Ernesto Hernandez-Umana is in US illegally after lying to border patrol when he entered in 2007

The captain of a boat that capsized in the New York harbor, resulting in the death of a woman and her infant daughter, has been taken into custody by Immigration and Customs Enforcement (ICE), with the federal agency saying it intends to deport him.

The deadly incident unfolded near the Statue of Liberty on 8 August. Authorities said a 911 call was received at approximately 10.25pm that prompted a major emergency response including multiple boat teams, a scuba team and a helicopter unit.

У Алексея Толстого в «Золотом ключике, или Приключениях Буратино» есть сцена, которая неожиданно хорошо описывает значительную часть того, что сегодня происходит вокруг больших языковых моделей.

В каморке папы Карло висит старый холст с нарисованным очагом. Огонь выглядит убедительно, котёл на месте, сама картинка создаёт ощущение домашнего тепла и порядка. На такой очаг можно смотреть и почти верить, что перед тобой настоящая печь.

Потом появляется Буратино, протыкает холст своим длинным носом — и обнаруживается, что очаг существует только на поверхности. За ним находится совсем другая реальность, а где-то дальше скрыта маленькая дверь, которую ещё предстоит открыть золотым ключиком.

Мне кажется, мы сейчас примерно в такой точке отношений с LLM.

Большая языковая модель за несколько минут способна сделать то, на что раньше уходили дни работы аналитиков, архитекторов и дизайнеров. Она нарисует архитектуру предприятия, бизнес-процесс, карту функций, модель управления, структуру исследования, концепцию продукта, требования к ERP, презентацию руководству. Всё получится аккуратно, современно и настолько убедительно, что заказчик совершенно естественно может посмотреть на результат и сказать: «Вот оно. Наконец-то всё стало понятно».

Но в серьёзной аналитике после первого впечатления должен прозвучать другой вопрос: что находится за этим холстом?

Некоторое время назад я писал здесь про антиспам для WordPress — историю о том, как из недоступного клиентам Cloudflare и дырявых бесплатных решений родился свой плагин геоблокировки, а чуть позже — вторую часть с разбором критики и рассказом, что успело поменяться. В конце самой первой статьи была одна проходная фраза: «параллельно ведется разработка аналогичного решения для 1С Битрикс — задача там несколько сложнее архитектурно, но подход тот же». С тех пор модуль доехал до Маркетплейса, обзавелся парой обновлений и первыми покупателями. Расскажу, что получилось — и почему для Битрикса такая защита понадобилась даже раньше, чем для WordPress.

• First Test day 2: Pakistan 171; England 366 for 8

• Brook misses out on home ton as hosts slip from 322-5

The clock had just ticked past 7pm, there was still half an hour of play left on this soggy second day, and it looked for all money that Harry Brook would end a year-long wait for his 11th Test century with a first at his home ground of Headingley.

Brook had agonisingly missed out by one run against India here last summer. And this time, against a ragged Pakistan side that had handed him a life second ball amid another outing to forget, the box was practically begging to be ticked.