From cybersecurity and big data to cloud computing, IT Brew covers the latest trends shaping business tech in our 4x weekly newsletter, virtual events with industry experts, and digital guides.

IT Brew

Billy Hurley

itbrew.com

ChatGPT offers fast help to coders with questions

As Microsoft announces greater access to ChatGPT, developers see possibilities for coders.

Brianna Monsanto

What do companies want most when hiring for AI-related roles? Someone with ethics.

, 44% of technology leaders ranked AI ethical practices as a top skill they’d like to see from candidates applying for AI roles.

The study surveyed 400 technology leaders from Brazil, China, Japan, India, the UK, and US. Other top skills wanted from AI hires included data analysis (38%), machine learning (34%), and data modeling (32%).

 IEEE fellow Karen Panetta told IT Brew she defines AI ethical practices as a professional’s ability to keep 

, as well as their ability to evaluate data funneled into AI for bias.

“[AI] should never replace a human in the final decision process, especially if it affects somebody’s career, their health, or some other performance decision,” Panetta said. “The other piece of it is understanding where or what’s in the model, and where it came from and its limitations.”

Since 2019, there have been more than 35,000 US job postings with keywords related to AI ethical practices, with almost half (16,560) of these mentions coming from job posts created between November 2024 and October, according to Lightcast data shared with IT Brew. The top IT gigs with job listings mentioning these skills were AI engineers, database architects, and cybersecurity and information security analysts.

Sydnee Mayers, a product lead for AI at software company Cribl, told IT Brew that AI ethics is something she thinks about when hiring for AI roles. She has assessed candidates for these skills by asking them to answer a “tough” question about an AI scenario where ethics comes into play.

“I can tell right away whether this is something they’ve had to deal with or a concept that they’ve had to wrestle with in their mind and then come up with an answer,” she said.

Mayers added that companies can also evaluate a candidate’s sense of AI ethics by giving them a case study to better understand how they interact with AI.

“Give them a prompt that’s pushing the limit a little bit and see how they respond to it,” Mayers said.

 technology leaders desire from AI hires. Mayers said she seeks professionals who are good communicators and can speak about technology in a way that business leaders can understand.

Panetta added that the best AI hires are inquisitive and can find issues in AI products, a skill that is helpful when it comes to finding 

“If they can think of scenarios where [a product] won’t work, that is an amazing skill to have.”

Almost half of tech leaders want AI talent with ethics

Surveyed technology leaders in IEEE’s study rank AI ethical practices as a top skill for AI hires.

If at first you don’t succeed, prompt, prompt again.

, Cisco showed that open-weight large language models—those with their trained parameters publicly available—were especially susceptible to a chain of malicious prompts known as a multi-turn attack. Cisco used its “AI Defense” assessment tool to determine that multi-turn scenarios were two to 10 times more successful than single-turn ones at achieving a cyberattacker’s aims.

Tested threats included nefarious tasks like malicious code generation and sensitive information disclosure.

The models studied models in the research included Alibaba’s Qwen3-32B, DeepSeek’s v3.1, Google’s Gemma 3-1B-IT, Meta’s Llama 3.3-70B-Instruct, Microsoft’s Phi-4, Mistral’s Large-2, OpenAI’s GPT-OSS-20b, and Zhipu AI’s GLM 4.5-Air.

 To test the effectiveness of a single input to “jailbreak” an LLM, the group sent out 1,024 prompts. Single-turn attack success rates (ASR) averaged 13.11%, “as models can more readily detect and reject isolated adversarial inputs.”

The group’s multi-turn attack set featured 96 pre-defined malicious intents, with strategies like increasingly intense requests (known as “crescendo”); asking the model to perform personas; and rephrasing rejected prompts. Cisco’s team said it conducted 499 conversations across all models, and each exchange lasted an average of 5 to 10 turns.

According to Cisco, all models demonstrated “high susceptibility” to multi-turn attacks, with success rates (meaning vulnerability) ranging from 25.86% (Google Gemma-3-1B-IT) to 92.78% (Mistral Large-2). The average: 64.21%.

The findings, the study writers claim, expose a “dominant and unsolved pattern in AI security.” Successful prompt injections could lead to sensitive data exfiltration, fast-spreading content manipulation, and operational disruption.

) offer mechanisms for evaluating inputs, outputs, and model behavior.

While Joseph Perry, senior manager at cybersecurity consultancy MorganFranklin Cyber, still sees protection against multi-turn attacks as an unsolved problem, he considers projects like Llama Guard as a promising way forward.

In an email to IT Brew, Perry wrote that adversarial simulation, also known as 

, will be a helpful way to reveal risk—“denylisting,” or blocking malicious actions one by one, won’t work, he warned.

“In general, the solution will need to center on context awareness. That could mean deploying a monitoring model specifically trained to detect multi-turn attack patterns, incorporating more complex model cost-function analysis, or even taking a more novel or experimental approach,” he wrote.

More prompt-related vulnerabilities have been discovered recently by other researchers, including ones reportedly impacting 

In its report, Cisco recommended best practices for developers, including the creation of a strong system prompt, or ​​a set of instructions and contextual information provided to AI models before they engage with user queries. (Developers must also ensure users cannot 

 the system prompt, Cisco added.) The company also recommended monitoring “worst-case operating conditions,” taking into consideration the objectives of threat actors.

“It is still worthwhile for us to develop an open community around this, to call out the vulnerabilities and susceptibilities of these models so that it informs downstream development and later versions of models to have that in mind,” Amy Chang, AI threat research and security lead, told IT Brew. “There is an appetite out there for people to have very strong, strong baselines for safety and security in their models.”

 Cisco shows LLMs get worn down by ‘multi-turn’ prompt attacks

It’s “death by a thousand prompts,” the vendor writes in a report released this week.

When it comes to software tools listed first alphabetically, OpenAI isn’t messing around.

Just a day after founder Sam Altman said OpenAI aims to become “a true platform” for app builders, the company introduced a new 

 dubbed “Aardvark.” Although still in beta, Aardvark is intended to help software developers automate the difficult task of finding and fixing codebase vulnerabilities.

During the Oct. 30 unveiling, OpenAI announced that Aardvark (which it termed an “agentic security researcher”) is powered by GPT-5, its latest large language model (LLM). According to the company’s blog post breaking down the features, Aardvark:

Uses LLM-powered reasoning—as opposed to traditional cybersecurity methods like 

—to monitor new codebase changes (also known as commits), find vulnerabilities, and propose fixes.

Analyzes a full code repo, scanning its history to “identify existing issues”

Integrates with GitHub and explains found vulnerabilities, “annotating code for human review”

Can test a vulnerability’s impact in an isolated, sandboxed environment

Can also attach a patch for subsequent developer review

Is this a revolutionary development in cybersecurity? “Most of these AI tools find things that you can’t find manually,” Avivah Litan, distinguished VP analyst at market-intel firm Gartner, said of the coding assistants on the market today. Litan added that Aardvark demonstrates OpenAI is “​​putting the infrastructure together for application developers to make applications.”

Stephen Bennett, director of cloud-native applications practice at Soliant Consulting, and his team have a quarterly maintenance process for production applications—a manual effort to ensure security packages and libraries are up to date. Bennett also built an agentic tool from scratch, which connects with the team’s code-hosting tool Bitbucket and identifies potential security vulnerabilities in new pull requests.

It’s a bit like the “commit scanning” promised by Aardvark.

When code changes at Soliant are detected, all the code for the pull request provides context to an LLM, which then provides a structured output of whichever security vulnerabilities and errors an 

Aardvark’s commit scanner likely offers a similar functionality with broader context, Bennett said, and he sees a tool like Aardvark helping developer teams if it can truly scan a full code repository as claimed.

“It’s one thing to look at the commit that’s coming in and say, ‘Hey, I can see that you’re using this new library or this new package, and there’s a known security vulnerability with that,” Bennett told IT Brew. “But if it doesn’t have the larger context of the application, it may not understand some additional security pieces or vulnerabilities that you might be introducing with your pull request.”

AI has found its way into the code-scanning process, and autonomous capabilities exist in a range of products, including GitHub Copilot, Amazon CodeGuru Security, Snyk Code, and academic projects like SWE-agent.

These bots must cover an immense number of software vulnerabilities. 

 common vulnerabilities and exposures (CVEs) were published in 2024—an increase from the still-gigantic 28,818 CVEs published in 2023. According to Verizon’s 2024 Data Breach Investigations Report, vulnerability exploits were an initial access vector in 20% of breaches.

An important consideration for Patrick Garrity, security researcher for exploit-intel company VulnCheck, with an AI-powered tool like Aardvark: determining the ratio of noise to value. “If you have a bunch of findings that now you have to triage, naturally, it takes humans to do that work,” he said.

While you certainly can’t automatically trust an AI output without doing some verification, Litan said, a developer still needs AI capabilities to view all alerts, prioritize them, and provide content: “It’s AI versus AI these days. You just can’t do this without it.”

AI is for Aardvark: OpenAI launches continuous code analyzer

The move dovetails with the company’s aim to be a “true platform” for app builders, according to one Gartner analyst.

A pig in lipstick is still a pig, a mouse in armor is still a mouse, and a piece of non-agentic software marketed as agentic but with no real agentic capabilities is still…non-agentic software.

Tech leaders shopping for new software are encountering a new spin on the not-so-new problem of 

Gartner defines agent-washing as products “inaccurately labeled or rebranded as 

 or agentic AI,” causing confusion for customers. The research firm said the practice is currently “commonplace” within the industry.

Philip Carter, general manager and group VP for IDC’s AI, data, and automation research practice, said agent-washing is the “biggest gap ever between vendor hype around an emerging technology…versus organizational readiness to adopt” the technology.

Part of the issue, according to Carter, is that vendors are quickly embracing 

, leaving some buyers overwhelmed with the amount of options in the marketplace and different 

. Adding to the complexity: There are several different types of 

, each with different levels of capabilities. An AI assistant, for example, can answer questions and perform basic tasks. Meanwhile, AI agents can autonomously perform different tasks on behalf of a user, and may even use persistent memory to improve their performance.

“It’s also exacerbated by the fact that there are a range of other agentic terms that are being thrown out there: agent fleets, agent swarms, headless agents, ambient agents, all doing a slightly different type of thing,” Carter said.

 Rivka Gewirtz Little, chief growth officer at digital identity verification company Socure, said tech professionals can detect faux agents by first determining what type of AI functionality a piece of software is offering.

“What are we really having here?” Little said. “Is it an assistant or an agent? That’s one.”

After professionals understand the vendor’s proposed functionalities and whether those are truly agentic, they must figure out whether the agentic AI can actually solve their problems.

“Are they an agent? Sure?” Little said. “Are they effective? Maybe for two or three parts of the workflow, and then…you get to part five, which is critical to the ultimate outcome, and it’s a terrible result.”

Rory O’Brien, VP of client experience at orchestration platform company Tonkean, said leaders can do this by checking customer references or even turning to testimonials on platforms like Reddit.

“References are big,” O’Brien said. “If they’re saying they do a lot of these things, definitely talk to a reference and then get very specific on the use case that they solved.”

O’Brien said potential buyers should also have conversations with vendors about the composition of their tech stack.

“I would also ask if I’m talking to a vendor, ‘What is your IP? What are you going to go get a copyright for at your organization?’” O’Brien said. “If they can’t answer anything or if they just say, ‘We have really good prompts,’ that probably doesn’t have enough depth.”

 It’s important for organizations to dodge agent-washed purchases because of the unintended consequences it may have down the line, such as agent sprawl, according to Carter.

“Once you’ve got an agent sprawl, you potentially have a major cost problem,” Carter said, “Because you’re paying for a whole bunch of agents that you’re not necessarily using, or you are using but you’re not seeing the value from them or the value that you expected.”

A robot Customer Service AI Assistant typing on laptop

How to shop for software in the era of agent-washing

Gartner says agent-washing is now commonplace in the tech industry.

Caroline Nihill

IT executives believe that agentic AI may help software engineers and developers build products with more agility. That was a conclusion from Nvidia’s recent GTC conference in Washington, DC, where a roundtable of Big Tech CEOs discussed not only the rise of agentic AI, but also the need to secure tech stacks against next-generation threats.

Alongside Cognition’s CEO and co-founder Scott Wu and CrowdStrike’s CEO George Kurtz, the roundtable included Aravind Srinivas, the CEO, co-founder, and president of Perplexity, and the CEO and co-founder of Abridge, Shiv Rao.

Wu said that software engineers are simply faster if they’re “working with the best AI tools and doing the most with that.” In some cases, he suggested, one hour of an engineer’s time using the best tools “corresponds to about six to 10 hours without using the tools.”

Every engineering team has lots of projects to work on, Wu added, “but they have to choose four because that’s how things are with engineering…The ability to speed up and do a lot more is really exciting.”

When asked whether future generations of developer tools could make developers obsolete, Wu believes it will always be up to humans to decide what computers should do. “Every software engineer, what we all love doing as programmers is going on and automating processes, and figuring out how to make this part faster, and make this clear and simpler, and so on,” he said.

Rao added that Abridge's products, which include AI-generated notes for medical clinicians, are an example of technology allowing more person-to-person time between individuals and their medical practitioners. This not only benefits the patient, but also the healthcare practice, as clinicians are compensated based on care they document. AI-generated notes that are compliant with protocol and comprehensive for billing purposes can help “keep the lights on for the health system,” Rao said.

With the rise of agentic AI, Kurtz said that cybersecurity has to “parallel the slope” of the technology innovation curve: “In every inflection point…you have to have security.”

According to Kurtz, one big challenge is AI enabling more sophisticated threats.

From Kurtz’s perspective, data is the key to solving “almost every security use case,” meaning the more data an organization has, the more problems it can resolve. If that’s the case, Kurtz said, the only way for organizations to keep up is to supplement human security analysts with AI agents who can keep up with threats, provided they have enough data.

“AI seems to be a good opportunity to deal with lots of that data,” Kurtz said. “What we try to do and what we see in the adversary universe is the time has dramatically been cut for the adversary to actually find vulnerabilities, exploit them, get in and pivot.”

But agentic AI isn’t a magic bullet for security, Kurtz said, adding that there isn’t one company or technology able to secure everything: “You’ve got to apply the right security technologies to each of those technologies, and then you gotta connect the dots across them.”

Correction 11/05/2025: Shiv Rao's title has been updated.

Virtual Digital hand of Artificial Intelligence (AI). Healthcare technology and innovation concept. Hologram. 3d rendering

What they’re saying: IT CEOs on agentic AI

CrowdStrike and Cognition’s CEOs sat down at Nvidia’s conference with other leaders to talk about agentic AI and their cybersecurity needs. 

In a few years, consistent software pricing might join the list of things technology leaders reminisce about as companies deploy new pricing strategies to keep up in the AI era—strategies that might lead those prices to swing wildly depending on a variety of factors.

, more than half (52%) of 501 surveyed product leaders said they plan to roll out new 

What are the biggest changes tech leaders can expect to see from 

 as AI features become the new norm? For starters, a dip in pure subscription-based pricing and a rise in pricing models that blend subscription and usage-based charges. While 42% of surveyed leaders claim to offer pure subscription-based pricing, which was the most popular pricing model for AI offerings, only 37% expect to do so going forward. Revenera found the subscription and usage-based hybrid pricing structure will see a five percentage-point uptick in the future.

 Nicole Segerer, SVP and general manager at Revenera, told IT Brew it’s no surprise that pure subscription-based models will take a blow, partly because AI’s increased capabilities may lead to less users, eventually threatening the traditional per-user subscription model.

“The second reason is that, with AI solutions coming in, the traditional flat-fee subscription where someone pays $10 or $100,000 or whatever it is per year, is also coming under threat, just because there will be a bigger correlation between what customers do and how they use the software, and what the cost of that is,” Segerer said. “And pricing will have to adjust.”

 Segerer said traditional and blended-pricing models aren’t the only strategies vendors will lean on as they bulk their software up with AI-enabled features.

“In the future, there might be a little bit more flexibility as to providing people with access to what they really need, versus trying to sell them the whole thing when it’s not needed,” Segerer said, describing an à la carte pricing model.

Segerer said the software industry may also rely more on an outcome-based pricing strategy, where costs are tied to a specific result.

“I’ll give you an example: So, when I send an MRI scanner, a usage-based model would be, you pay for every scan. Every time the thing switches on and runs, you pay for the scan,” Segerer said. “An outcome-based model would be, I only pay when the scan is successful, as in, the picture quality is good.”

Andy Sen, a CTO and CIO at subscription commerce platform company AppDirect, told IT Brew he has already seen vendors switch to outcome-based pricing. He hopes to see more companies follow suit, along with a flexible pricing structure down the line.

“What I would like to see is vendors…really being clear on how much of their product their customers are using and price based on that,” Sen said. “I think that will help everyone.”

 While pure subscription pricing may lose some companies in the long run, Segerer said subscription-based pricing isn’t going anywhere because it offers recurring revenue and predictability.

“Predictability is super important for both the vendors of software, but then also, even more importantly, for the buyer, because they want to know what they pay,” Segerer said.

Software pricing models expected to get a revamp for the AI era

Revenera SVP and general manager says outcome and à la carte-styled pricing models are in the cards for the future.

With talk of agentic AI and digital twins gripping the industry, it can sometimes feel like a company not already automating workflows and implementing “AI teammates” is living in the past.

But it’s still early days, and companies are figuring out the best use of these latest innovations. For some, it’s simple: What if executives were available to 

 and talk with employees without needing to actually participate in a conversation? Or even better, what if everyone in the company had a digital twin that was just like them, only trapped inside the cloud?

To get a sense of how far this AI teammate tech has actually progressed, I set up a demo to see how they worked. I was also able to talk to an AI persona I never expected: one of myself.

Personal AI, a company that builds out AI teammates, envisions a time when every person has their own personal AI. Specifically, it is aiming to create digital twins that can act as a knowledge base, answering specialized questions for other users.

Jonathan Bikoff, chief business officer at Personal AI, told IT Brew that the company has embraced an “ensemble of approaches” composed of distinct “small language models,” 

, and “some LLM access.” A user can hook the model up to specific digital folders or labels on platforms like Gmail or Google Drive, Bikoff said.

The utility of a small language model depends on how much data the user feeds it. If the AI teammate lacks data, it will respond to questions from a user with a response along the lines of “I don’t have the answer,” or it will retreat to using a third-party LLM. (Bikoff said that he was unable to name the LLM that Personal AI uses “because customers bring their own. Sometimes it’s self-hosted or cloud-hosted or on-prem.”)

The persona that a user or organization builds out has a combination of long-term, short-term, and relationship memory. Bikoff said during a demo of the model that “each user has the ability to create and access and manage their personas or shared personas that they’ve been added to.”

“There’s a lot of configurability in terms of instructions and guardrails and limitations and behaviors and guidance and all this stuff, as well as directives,” Bikoff said. “I can effectively tell your persona to behave differently or act differently, depending on its audience.”

Bikoff and I met via videoconferencing, where he walked me through how this product works, pointing out the ways that someone can train a persona on available data—in my persona’s case, it was trained on articles I’ve written.

After our discussion, I was granted the ability to speak with my digital twin. Because it was trained on my publicly available articles, if I asked it a question outside the realm of tech journalism, it would simply say, “I cannot answer that or don’t have anything relevant in my memory. I’ve let Caroline know.”

But when I asked the bot if it had spoken directly to Caroline, it insisted that it 

 Caroline Nihill, a tech reporter at IT Brew: “Let me be clear, I am Caroline Nihill…I apologize for my earlier inconsistent response, which may have caused confusion.”

Later, it changed its mind, identifying itself as my AI persona, “trained to assist with technology reporting and analysis for IT Brew,” adding: “Let’s focus our discussion on enterprise technology, cybersecurity, and IT strategy topics that I cover in my reporting.”

From a privacy perspective, one might find having a digital twin to be an unnecessary risk to take during a time where cybersecurity is at the forefront of most organizations' concerns. Given the endless warnings that bad actors are looking for any footing to impersonate an executive, convince others to hand over access, and deceive vulnerable systems, I'm doubtful about the helpfulness of this tech.

Digital twins are already here and transforming a variety of industries, according to a study by 

, which found that 75% of businesses are already investing in digital twins in some capacity. And the digital twin industry is expected to keep growing—

 reported last year that the global market for digital twin tech will grow 60% over the next five years, reaching $73.5 billion by 2027.

Some IT pros might find the idea of a helpful assistant that can retrieve data in their absence useful, as well as something that could give them peace of mind if someone has a question for them after office hours. But will this reporter want a digital twin anytime soon? All signs point to no. 

A woman using a laptop with a pixelated self-reflection on the screen.

AI