By Adam Mills
AI Editor
0
Share
Newsweek is a Trust Project member
See more of our trusted coverage when you search.Prefer Newsweek on Googleto see more of our trusted coverage when you search.

Newsweek
This is AI Impact, Newsweek's weekly newsletter where each week, we will explore how business leaders are unlocking real value through artificial intelligence.
Read More on AI
Tap or click here to get this newsletter delivered to your inbox.
Editor's Note

Newsweek
Before we get into this week’s edition, I want to invite you to join us for an upcoming Newsweek webinar presented by Cognizant: “Humanizing AI: Personalization, Engagement and the Customer Experience.”
On Thursday, July 30, at 2 p.m. Eastern, Gabriel Snyder, Newsweek’s executive editor, enterprise, will lead a discussion on how companies are using AI across service, support, product experience and customer engagement while balancing automation with trust, transparency and human judgment.
Nick Mehta, entrepreneur in residence at Bessemer Venture Partners and former CEO of Gainsight, will join the conversation, along with an executive from Zendesk and additional guests to be announced.
Registration is free, and I’d love to see you there.
Signal Capture
Signals from the frontlines of AI adoption
AI Agents Could Remove a Human Security Checkpoint, Menlo Security CEO Warns
By Adam Mills
An invoice submitted for payment can look ordinary to an employee—even while containing hidden instructions that tell an AI agent to redirect the money.
Menlo Security, a cybersecurity company focused on browser security, has described a scenario in which the command is concealed in white-on-white text. A person reviewing the invoice would not see it. An agent could interpret it as part of the task it was authorized to perform.
Bill Robbins, CEO of Menlo Security, said the risk grows when software moves from advising a person to acting without human approval at each step.
“When an AI agent becomes an active user, that checkpoint disappears,” Robbins told Newsweek.
Recent announcements from OpenAI and Anthropic show how quickly AI’s cybersecurity capabilities are advancing. OpenAI says Daybreak is designed to help defenders find and validate vulnerabilities, test patches and verify remediation. Anthropic reports that Mythos Preview represented a substantial leap in vulnerability discovery and exploitation, and initially released the model to a limited group of critical industry partners and open-source developers.
“Right now, these models are new and their capabilities are genuinely dual-use,” Robbins said. Anthropic rolled out Mythos selectively, he said, because “the same capabilities that are invaluable for defenders are also extremely dangerous in the wrong hands.”
Inside enterprise workflows, material an agent is authorized to process can still contain commands the organization never approved. Security researchers call this prompt injection: malicious instructions placed in a web page, document or other content that can redirect an agent’s behavior.
“Prompt injection is the most urgent risk because unlike traditional malware, it doesn’t require a code vulnerability,” Robbins said.
A manipulated agent may also have permission to access multiple applications and sensitive information. Menlo’s “2026 State of Browser Security Threat Report” describes machine-speed data exfiltration as a distinct risk because a compromised agent can transfer data at a scale a human attacker could not match.
The browser matters because an agent can encounter a malicious instruction and act on it in the same session. Robbins said traditional security tools may monitor traffic moving to and from a device or activity on the device itself without showing exactly what the agent read or did inside the browser.
From Jan. 1 through March 31, Menlo said its platform blocked 52,185 threats originating from sites already classified as safe. The figure covers browser threats such as phishing and malware delivery, rather than attacks on AI agents specifically, but it illustrates the limits of relying on a site’s reputation to determine whether its content can be trusted.
Robbins said one requirement for governing agents is separating the content they are allowed to analyze from the instructions they are authorized to follow.
“Human judgment can’t be removed, but instead has to move upstream,” he said. Analysts still need to define guardrails, limit agents to the access required for a specific task and review forensic evidence when something breaks.
Robbins said that work “creates the conditions under which agents can actually be trusted to operate autonomously.”
Core Intelligence

...
Cognizant Sees AI Shortening the Path to Expertise
Cognizant’s plan to hire 1,500 U.S. graduates treats AI as a way to move entry-level employees from onboarding to meaningful client work sooner.
Roughly 600 of the planned hires have already joined the company, CEO Ravi Kumar S said. The broader hiring effort spans Cognizant’s technology services business, its Belcan engineering subsidiary and the Frontier Engineers program. AI is expected to help new employees reach substantive assignments sooner, compressing a development path once built around years of routine work.
The hiring profile is broad, with candidates coming from technical and non-STEM disciplines.
“Writing software has become relatively easy,” Kumar told Newsweek.
When code is easier to produce, the scarce skill becomes knowing which problem to solve and how the work actually functions. A legal process requires knowledge of law. AI applied to finance, health care operations or employee onboarding still depends on people who understand the underlying decisions, exceptions and constraints.
“We’re looking for curiosity, problem finding, problem solving, understanding a specific domain or a specific function,” Kumar said.
Cognizant’s approach depends on building that domain knowledge quickly. New hires receive eight to 12 weeks of training intended to prepare them for client work. From there, some graduates may be deployed to client sites alongside experienced employees, where they can learn how a company operates and absorb practices and institutional knowledge that may not appear in formal process documents.
Kumar does not treat workforce development as a separate human resources program.
“We think training should be a part of the AI infrastructure,” he said.
The logic also cuts against the idea that entry-level hiring becomes optional once AI absorbs more junior tasks.
Companies still need a pipeline that produces experienced specialists, client advisers and managers. Early-career employees eventually become the people expected to understand clients, coach teams and take responsibility for work performed by people and agents.
Experience still has to come from somewhere, even when AI handles more of the production work. A shorter apprenticeship still has to develop the judgment and domain knowledge that the next generation of specialists and leaders will need.
You can read the full article here: Cognizant Is Hiring 1,500 Graduates for an AI-Driven Workforce.
Upcoming Webinar
Is India on the Right Side of the AI Trade?

India’s role in the global technology economy has long been shaped by scale: a massive IT services sector, fast-growing capital markets, digital public infrastructure and a young workforce. AI could raise the stakes even further by changing where value is created, who captures it and how countries compete.
In an upcoming “AI Impact Forum” session, Dr. Ranjit Tinaikar speaks with Shri Ashishkumar Chauhan, CEO of the National Stock Exchange of India, about whether India is positioned to benefit from the AI revolution—and what the technology could mean for the country’s economy, financial markets, market infrastructure and next stage of growth.
Join the live discussion on Thursday, July 23, at 10 a.m. Eastern. Register for free.
Prompt Injection
What’s one recent insight you’ve learned about AI?

...
“After we filmed Martin Scorsese's class, I saw the colored cut and hated it. I remember being so upset and asking my team: how did this happen?! They said: ‘David, Martin did the color.’ I felt like an idiot and had to recalibrate.
I came to call this the Scorsese Test. If someone can do something 1000x better than me, why am I doing it? If I use AI to make a film, it will never be better than Scorsese using AI. So why compete? The better question is: what could I make that Martin Scorsese simply can't? He can't make 100,000 personalized films. No person can. If someone world-class can do it better, you're being incremental. The real opportunity is to create what wasn't previously possible at all.
It changed the questions I ask. Before, I was focused on how AI could make what we already do faster or better. Now I start every conversation with: what does this make possible that couldn't exist before? That shift sounds subtle. It isn't. One question leads to incrementalism. The other leads to entirely new categories. At MasterClass, that means personalized learning at a scale no human instructor could ever reach.”
Have your own lesson to share? Email us at: a.mills@newsweek.com
Run Log
AI use case of the week
Akido Takes AI Beyond the Clinic

...
Street medicine starts from a basic reality: many people living outside cannot reliably get to a clinic. They may lack a phone, transportation or stable housing, even when their medical needs are urgent.
Akido, a health care company that combines clinical AI with care delivery, began with street medicine, which brings care directly to people living outside. Patients may have several chronic conditions, behavioral health needs and no consistent connection to the health system.
Prashant Samant, chief executive officer of Akido, said clinician shortages make the traditional visit difficult to extend to more patients. Doctors generally take a patient’s medical history, ask follow-up questions, consider possible diagnoses, create a plan and document the encounter one step at a time.
Akido built ScopeAI to take on parts of that work before a clinician makes the final decisions. During a visit, a medical assistant or community health worker stays with the patient while ScopeAI guides intake in real time, asks follow-up questions, considers possible diagnoses and drafts an assessment and care plan. A licensed clinician reviews the work and remains responsible for every final decision.
According to Samant, Akido has served thousands of patients through the model, often providing care the same day rather than requiring them to schedule an appointment, arrange transportation and travel to a clinic.
“The emergency room is often where the system finally catches people after it failed them upstream,” Samant said.
The street-medicine program is financially sustainable through Medicaid rather than dependent on a charitable or grant-funded pilot, he said.
Akido began with street medicine to see whether the care model could work in one of health care’s hardest settings.
“The hardest patients are the best proof,” Samant said.
Akido’s model brings the visit to the patient. A community health worker stays at the patient’s side while AI prepares the clinical information for a licensed clinician to review and decide what care comes next. Earlier contact gives the health system a chance to reach patients before they arrive at the emergency room.
Have an interesting AI use case to share with us? Email us at: a.mills@newsweek.com
Context Window
■ An Exponential View analysis estimates that global AI sales outside China reached $25 billion for hyperscalers and neoclouds in the first quarter, exceeding their estimated depreciation costs for a second straight quarter but leaving little room for power, labor and financing expenses. [Los Angeles Times]
■ Rising AI costs are pushing some U.S. companies toward cheaper Chinese open-weight models, with Lindy.ai saying its move from Anthropic to DeepSeek-V4 saved the startup millions of dollars. [NPR]
■ A Bain survey found that 86 percent of industrial CEOs are prioritizing AI for cost reduction and productivity, even as 90 percent say their programs are underdelivering because of capability gaps, stalled pilots and uncertain returns. [Bain & Company]
■ New York Gov. Kathy Hochul said her administration used AI to review state rules, regulations and policies for outdated provisions, completing in months a process she said could have taken staff five years. [The Verge]
■ Nvidia introduced Cosmos 3 Edge, a world model designed to help robots and vision agents navigate physical environments, and said Fujitsu, Hitachi and Kawasaki Heavy Industries intend to join a new physical AI coalition in Japan. [CNBC]
■ Oracle introduced a new builder experience for AI Agent Studio that lets customers and partners create agentic applications inside Fusion Applications, with no-code and pro-code tools plus built-in security, approvals, governance and audit trails. [Oracle]
Transfer Protocol
Tracking executive moves across the AI landscape
Gil Regev has been named executive vice president and general manager of AI at Mend.io, where he will lead strategy and execution for Mend AI across AI supply chain visibility, behavioral testing and runtime protection.
Swamy Ramajayam, previously director and head of people data strategy, architecture and AI foundations at Meta, has been appointed Mariner Finance’s first chief AI and data officer, leading enterprise data and AI strategy, data infrastructure modernization and responsible AI adoption across the business.
Yuriy Demidko, previously chief information officer at DP Fox Ventures, has joined Clarivoy as chief technology and product officer, overseeing technology strategy, product vision, engineering, data infrastructure and artificial intelligence initiatives.
Ron Dovich, previously senior vice president of engineering at Brinqa, has moved into the role of chief AI and automation officer, with responsibility for the cybersecurity company’s AI and automation strategy.
Faraz Iqbal, previously chief technology officer at Edible Brands, has been named chief technology and AI officer at BBG, where he will lead enterprise technology strategy, AI and data transformation, technology modernization and the scaling of global operations.
Roger Maeda, previously director of enterprise applications and application development at Arnold & Porter, has become the firm’s first chief artificial intelligence officer, leading a dedicated team that will build, test and deploy AI tools in support of client work.
Know someone on the move in AI? Send job change info to a.mills@newsweek.com
Magic Moment
What’s the most fun or unexpected way you’ve used AI lately?

...
“The most unexpected way I've used AI recently wasn't sitting at a desk. It was while running through Brooklyn. Before the trip, I built a 5K route around Notorious B.I.G. landmarks. During the run, I talked naturally with AI through my earbuds while wearing AI glasses that captured the experience.
Every mural, neighborhood, bridge, and subway station became an opportunity to ask another question. Why was this mural painted? Where did Biggie grow up? How did this neighborhood change over time? Instead of following a scripted audio tour, it felt like running with an incredibly knowledgeable local who could answer whatever I was curious about in real time.
That experience reminded me why AI has become part of how our family learns together. For years, I've used it with my kids whenever they ask questions about the world around them. Whether we're wondering why trees grow on old quarry hills, how bridges are built, or why something works the way it does, AI has become less about finding quick answers and more about exploring curiosity together. The best AI experiences don't make you stare at a screen longer. They make you pay more attention to the world around you.”
Experience some AI magic? Tell us about it at a.mills@newsweek.com
Tap or click here to get this newsletter delivered to your inbox.
View 0 Comments
Submit Correction
View Editorial & AI Guidelines