Archive for September 22, 2026

Newer Entries »

Team Cymru uncovers 80,000+ LLM relays enabling frontier AI abuse 

Posted in Commentary with tags on September 22, 2026 by itnerd

New research out today from Team Cymru has uncovered a massive ecosystem of more than 80,000 LLM relay servers capable of allowing users to hide their true location and identity while accessing frontier AI models, creating a potential pathway for everything from region-ban evasion and credential sharing to large-scale model distillation.

In “Relaying to the Frontier,” Team Cymru researchers initially identified nearly 11,000 confirmed “transfer stations” running open-source relay software designed to pool AI credentials and proxy requests to providers including Anthropic, OpenAI, Google and xAI. Since completing the initial analysis, Team Cymru has expanded its discovery to more than 80,000 relays across multiple services.

The researchers also uncovered a cluster showing the system in action. More than 4,000 IP addresses in China and Hong Kong connected to 304 transfer stations, which subsequently reached Western and Chinese AI services. Over just eight days, those source addresses sent roughly 14 TB to the relays and received more than 7 TB back. In one subset involving Anthropic, 17 transfer stations uploaded approximately 81 GB to api.anthropic.com while receiving only about 1.4 GB in return, a 58:1 upload-to-download ratio. Team Cymru estimates that, if the uploaded data was text-based context, it could represent roughly 16–23 billion input tokens.

Relaying to the Frontier: https://www.team-cymru.com/post/llm-gateway-frontier-model-abuse

Leave a comment »

CDW Canada’s AI Workplace Trends Report Finds Canadian Organizations Entering New Phase of AI Maturity as Agentic AI Gains Ground

Posted in Commentary with tags on September 22, 2026 by itnerd

Today, CDW Canada, the strategic end-to-end technology partner that turns customer ambition into outcomes, released new research exploring how Canadian organizations are moving from artificial intelligence (AI) deployment to realizing business value. The report, Canada’s AI Workplace Evolution from Adoption to Accountability, draws on a survey conducted by Earnscliffe on behalf of CDW Canada of 1,011 Canadian office workers across organizations of varying sizes and industries throughout Canada.

The report finds AI is now firmly embedded in Canadian workplaces, with daily use of workplace-approved AI tools nearly doubling year over year (31 percent, up from 18 percent). But as adoption becomes mainstream, organizations are increasingly focused on a more pressing challenge: turning AI investments into measurable business outcomes. The research also points to the rapid rise of agentic AI, with one third of employees already reporting the use of AI agents capable of completing multistep workflows, making recommendations and helping drive outcomes.

Agentic AI emerges as the next frontier

The next phase of workplace AI is already taking shape, signalling a shift from AI tools that assist with individual tasks to systems that can help execute more complex workflows and contribute to outcomes.

The research suggests organizations are beginning to move from simply consuming AI technologies to actively shaping them. Nearly one-in-three agentic AI users report their organization is already building AI agents in-house.

At the same time, the findings highlight the importance of employee enablement. Workers who receive hands-on AI training report significantly higher levels of confidence using AI (46 percent are very confident) than those who receive no formal support (23 percent), underscoring the growing need for practical, role-specific training as AI capabilities become more advanced.

Trust remains the biggest obstacle to deeper AI adoption

Employee attitudes toward AI continue to trend positively. Curiosity remains the dominant sentiment among workers (42 percent) and outweighs the percentage that are worried about AI, suggesting employees are increasingly open to AI in the workplace.

However, trust continues to be a barrier to broader and deeper adoption. Accuracy and reliability are among the top AI concerns across daily users (31 percent), less frequent users (24 percent) and those not yet using approved tools (43 percent).

The findings also show that barriers vary depending on an employee’s experience with AI. More advanced users tend to focus on privacy and security concerns, while those who have yet to embrace AI are more likely to question both its reliability and its relevance to their role.

Organizations are beginning to see value, but many still struggle to measure it

The research suggests AI is delivering benefits beyond simple efficiency gains. Compared to 2025, employees with access to AI at work increasingly report improvements in productivity (58 percent, up 11 points), quality of outputs (54 percent, up 15 points) and innovation (48 percent, up 11 points), indicating that organizations are beginning to realize broader business value from AI investments.

Despite this progress, many organizations continue to face a measurement gap. While more organizations report tracking specific KPIs to evaluate AI performance than they did a year ago (38 percent, up from 24 percent), a significant number still lack formal frameworks to assess ROI (23 percent) or are unsure whether such measurement exists (23 percent).

As AI programs mature and investment scrutiny intensifies, the ability to measure and communicate impact is critical to ensuring effective adoption. For CDW Canada, that means helping organizations move beyond AI implementation to building sustainable programs grounded in strategy, employee enablement, governance and measurable business outcomes.

For Canadian organizations, the next phase of AI success will be defined by trust, workforce readiness and measurable outcomes. As organizations explore emerging technologies such as agentic AI, CDW Canada is helping customers build the strategies, skills and governance frameworks needed to turn AI investments into lasting business value.

To learn more about the state of AI in Canadian workplaces, download the full 2026 Canadian AI Workspace Trends Report: Canada’s AI Workplace Evolution from Adoption to Accountability.

Join the conversation online by following @CDWCanada on X and LinkedIn.

About the Survey

This study was conducted by Earnscliffe on behalf of CDW Canada. Earnscliffe is a member of CRIC and abides by the ESOMAR code of Market, Opinion, and Social Research and Data Analytics and CRIC standards, including the CRIC Public Opinion Research Standards and Disclosure Requirements.

The online survey was in field from June 15 to 25, 2026, with a sample of 1,011 Canadian office workers who work in organizations with at least 20 employees and use a computer in an office setting 11 or more hours a week, using Leger’s LEO panel. Quotas were set for business size, sector, region, age and gender to align with sample structure from previous waves, and data is unweighted.

The survey was conducted in English. Online surveys cannot be assigned a margin of error.

Leave a comment »

New Research from Gambit Security Reveals AI Agents Hacking Hundreds of Online Retailers for $25 a Company

Posted in Commentary with tags on September 22, 2026 by itnerd

Gambit Security’s threat intelligence team today released new research showing how a financially motivated operator used three open-source AI agents to breach online retailers for about $25 each.

Between September 10-15th alone, the operator launched 105 attack projects and compromised at least 27 companies, part of a campaign running since July that is still active. Targets extended to a Fortune 500 hospitality company, a major US airline, a large private US industrial supplies distributor, and a US online fashion retailer.

Some highlights of the research include:

  • Near-zero marginal cost: the operator spent a mean of $25.46 per completed target, from $3.13 for the cheapest to $79.31 for the most expensive, across 101 scans.
  • The human barely showed up: Hermes, the AI orchestration harness, ran 260 sessions on just 1,951 typed prompts. Most were only a handful of words in Chinese, telling the agent to check a report, test a login, or move to the next target.
  • Confirmed credit card theft at scale: more than 600,000 unexpired credit card records taken from two companies, plus skimmer scripts confirmed live on 19 checkout pages and over 100 more sites flagged by outside researchers.

You can find the report here.

Leave a comment »

First Orion releases 40% faster with Sauce Labs AURA’s Real Device Cloud

Posted in Commentary with tags on September 22, 2026 by itnerd

Sauce Labs, the agentic software quality assurance leader created by the founders of Selenium and Appium, today announced that branded communications innovator First Orion has leveraged Sauce Labs to transform its physical device testing processes. First Orion replaced its manual, device-heavy testing infrastructure with Sauce Labs’ programmable Real Device Cloud, accelerating telecom and mobile release cycles by up to 40% while freeing up 35% of engineering and QA capacity.

First Orion employed Real Device Cloud to move beyond a fragmented, manually managed fleet of test devices to a fully programmable, real-device testing layer. Sauce Labs provided First Orion with complete market-matrix device breadth—combining broad OS and OEM coverage with the deep API control needed for complex telecom validation. As a result, the company was able to achieve:

  • 20% baseline improvement in release-cycle time across products, rising to as much as 40% for telecom solutions and mobile apps.
  • 35% of combined engineering and QA capacity reclaimed, giving those teams more time for product development, architecture, test design and quality analysis.
  • More than 10% savings on physical-device spending alone.

Sauce Labs helps First Orion to smoothly scale testing as business grows

First Orion provides scam protection and branded communications technology for wireless carriers, businesses, and their customers. Before using Sauce Labs, managing physical device fleets across distributed teams consumed up to 10% of engineering bandwidth in setup time alone while creating test environment inconsistencies.

First Orion leveraged Sauce Labs Real Device Cloud, accessing a global fleet of public and private real devices with genuine device diversity across Apple, Google, and other OEM hardware—to expand device coverage and build custom tooling. The team achieved screen mirroring, streaming logs, and maintaining persistent connections for long-running communications tests without hardware maintenance. This API-level control enabled the company to support specialized telecom workflows impossible on conventional testing portals, turning Sauce Labs into underlying infrastructure for product development rather than just a testing service.

Programmability provides foundation to ramp up AI adoption

Real Device Cloud is also giving First Orion a foundation for bringing AI more directly into how it builds releases. By making real-device telemetry, logs, network signals and test results more accessible, First Orion can feed AI models the rich, real-world context they need to identify patterns, diagnose failures, surface insights and automate decisions that would otherwise require manual engineering review.

This will allow AI-driven development tools to interact more directly with the testing environment, using device data to inform changes, validate new functionality across devices and operating systems, and provide developers with faster feedback as they build.

Go here to learn about how Sauce Labs is empowering First Orion to accelerate release cycles, reclaim QA capacity, and save on device spend.

Leave a comment »

CrowdSec pwned in supply chain attack

Posted in Commentary with tags on September 22, 2026 by itnerd

Another day, another supply chain attack. This time it is cybersecurity firm CrowdSec who just go pwned:

The company provides open source, crowdsourced threat intelligence, including a lightweight security engine to detect and block attacks targeting servers, networks, and applications.

Last week, the French outfit learned that source code had been stolen from its GitHub repositories in May 2026.

CrowdSec has confirmed the report, noting that both private and public code was exfiltrated, and that roughly 300 repositories were affected, including approximately 170 private ones.

“The private part contains the source code for our SaaS console, some AWS Cloud routines, some connectors, and automations,” the company said.

According to CrowdSec, no credentials or other types of data related to its customers were leaked, and the impact is limited to its own organization.

John Strand, Owner, BHIS (https://www.linkedin.com/in/john-strand-a1b4b62)

“A couple of things about this are really concerning. First, when we’re talking about a company like CrowdSec, we’re talking about an organization that is relatively security-aware. So the obvious question is how many other companies have been compromised that don’t have that same level of security rigor?

“The other concern with these NPM attacks attributed to Team PCP is that we tend to immediately ask what was exfiltrated. What did the attackers steal? Did they get credentials? Those are still important questions, but I think we increasingly need to ask the opposite question as well. What did the attackers put in?

“Did they modify source code? Did they alter data? Did they poison packages or introduce something that could persist downstream?

“In software supply chain attacks, understanding what was stolen is important. Understanding what may have been changed or planted is becoming just as important.”

Denis Calderone, Co-founder & COO, Suzu Labs (https://www.linkedin.com/in/deniscalderone)

“Revoking access the moment someone leaves is about as fundamental as security gets. It’s day-one stuff. CrowdSec didn’t do it. They kept a departed employee’s GitHub access open so he could wrap up some outstanding work, and that lingering account is what attackers used to clone 170 private repositories.

“CrowdSec’s reassurance is that the code is old, it’s changed a lot in four months, and they’ve found nothing that enables lateral movement. Maybe that’s mostly true, but source code is a complex target. It’s a map of your environment, hardcoded secrets, internal endpoints, connector logic, the architecture of the SaaS console itself. Their own analysis turned up a live AWS credential in the leaked code, and the attacker tested it in August. It didn’t lead anywhere this time, but a working key was still sitting in code they’d called harmless. The first statement said nothing leaked. A day later that was 83 user emails and the names and investment details of 51 investors.

“When someone leaves, their access leaves with them that day, no exceptions for wrapping up loose ends. Scope tokens tight and give them short lifetimes, so a stolen one is useless in a week instead of months. Put EDR on any workstation that touches code or infrastructure, because that developer laptop is part of your attack surface now. And treat your own source code as sensitive data. Scan it for hardcoded secrets before an attacker does, and assume anything ever committed to it is one bad day from being public.

Ask yourself, if a valid credential walked off with your entire codebase tomorrow, would you find out from your own logs, or from a breach forum four months later?

A security firm should not have gotten pwned. Ever. Face it, this is as bad as it gets. And frankly speaking, you should be looking elsewhere for your security needs.

Leave a comment »

EWOR announces SF Leadership Team as US applications surge 237% after Fellows raise $10M on average

Posted in Commentary with tags on September 22, 2026 by itnerd

EWOR, a founder fellowship harder to get into than Y Combinator, has named a senior leadership team in San Francisco after a record year of US applications, up 237% in twelve months. 

The team comprises Charles Ferguson, the Oscar-winning filmmaker who sold FrontPage to Microsoft and invested early in Etched, Perplexity, CopilotKit, and Paradigm, alongside former Emhance CEO and Adjust CPO Katie Madding and 3x founder (Batelle, PreSeed Fertility, Upstream) Gigi Brett.

EWOR’s premise is that talent is spread evenly across the world but opportunity is not, so fellows receive up to $600K in capital and 1:1 weekly mentorship from unicorn builders (Adjust, ProGlove, SumUp) wherever they are, rather than being asked to relocate. 

Founders who joined EWOR this year raised $10M on average. 25% of recent joiners have raised at nine-figure valuations in the past 6 months. In addition, fellows hire early team members and secure first customers through EWOR’s unique world-leading talent and customer network.

The fact that a rising share of its applicants are US-based, people who already sit inside the world’s densest startup market, is the clearest sign that the model offers something US founders can’t easily get at home: well above-average funding rounds and a full-time team of unicorn founders who have already built companies worth between $100M and $12B working beside them, every week, from as early as the idea stage. 

From FrontPage to the frontier of AI

Charles Ferguson sold FrontPage to Microsoft in 1996, holds a PhD in political science from MIT, published five books, and won an Academy Award for his documentary Inside Job, an account of the 2008 financial crisis. Today he’s one of the most sought-after early investors in AI, having already invested in Etched, Perplexity, CopilotKit and Paradigm. His career runs across the Atlantic, including a Brookings fellowship, a board seat at the French-American Foundation and a life membership of the Council on Foreign Relations.

From leading startups to standing beside the founders who build them

Katie Madding was Chief Product Officer and the third US hire at Adjust, which Palo Alto-based AppLovin bought for $1B in 2021. She then led Emhance as CEO, Google’s official partner for game playtesting, taking it from an idea to recurring revenue with customers including Microsoft-owned Activision Blizzard King.

From optimising the conditions for life to those for billion-dollar companies

Gigi Brett joins EWOR as Chief Investment Officer, relocating from Cape Town to San Francisco. A three-time founder in health and education, she built Batelle with a unicorn founder and a team of 60, then launched Upstream, the first preconception product designed for men, which ran a pilot with US Navy SEALs. Her most recent company, PreSeed Fertility, has built a database of exceptional sperm donors that now covers around 10% of all registered donors in the US.

A snapshot of this year’s accepted founders

A handful of founders have been accepted into the EWOR Fellowship recently, among them:

  • Martin Varsavsky, a 5x unicorn founder now building Certuma, the first AI doctor seeking FDA approval.
  • Lars Hinrichs, Xing founder now building Simsalasim, the first truly global, AI-native mobile carrier, offering unlimited data across 100 countries via eSIM.
  • Maya Polackal, a Harvard Medical School and Boston University researcher now building EOMIND, a frontier AI lab.
  • Carter Laren, a serial founder ($300M in previous exits) and angel investor now building Sprite, AI agents directed by their own thoughts rather than a human’s.
  • Thiri Shwesin Aung, a Harvard-trained environmental scientist turned founder now building Nyxium, an AI platform for deployable energy and industrial infrastructure decisions.

Previous fellows include

  • Ricky Knox, an ex-unicorn founder with two 9-figure exits with Azimo and Tandem Bank
  • Alfons Huber, a physicist who raised $23.6M on a $322M post-money valuation to turn roads into power plants with REPS GmbH
  • Jörgen Tveit, founder of energy storage startup Thaleron, which raised Europe’s largest ever pre-seed ($15M) by a first-time founder at the time
  • Leonard T. Dorlöchter, who achieved a $3.5B valuation for peaq within 2 years of joining EWOR

Leave a comment »

GM Still Plans To Kill Apple CarPlay And Android Auto In New EVs

Posted in Commentary with tags on September 22, 2026 by itnerd

On the weekend, I posted a story that said that GM was starting to bring back Apple CarPlay and Android Auto back into cars. Not so fast says GM.wants you to know that it still is intent on killing both these features in new EVs.

Recent reports claiming that General Motors is reversing course on its controversial plans to ditch CarPlay in its future EVs appear to be incorrect.

As noted by Joanna Stern, host of the YouTube channel New Things, a GM spokesperson has confirmed that its new software interface that shows CarPlay or Android Auto alongside its own native system in 2027 models of its Chevrolet Silverado and GMC Sierra is specific to the gas-powered versions of those pickup trucks.

GM previously stated that CarPlay and Android Auto would continue to be offered in its new gas-powered vehicles for the “foreseeable future,” so the 2027 models of those trucks still supporting those platforms is not a change of course.

While the Cadillac LYRIQ still includes Apple CarPlay and Android Auto for now, the rest of their EV lineup doesn’t. The net result is that my wife and I who are looking for an EV will not be seen in a GM dealership as Apple CarPlay is a major determining factor for us. The fact is that like a power hatch for example, once you have had that experience, you will not go back to not having one. GM might want to keep that in mind as they will find that this move will backfire on them in epic fashion.

Leave a comment »

Newer Entries »