The conversations enterprise IT leaders actually listen to.
Volume 342, Issue 2 · 20 articles from 4 issues
Fred Heiding of Menlo Park Intelligence explains how frontier AI models manipulate people and why that threat is hard to defend.
At the Dark Reading News Desk during Black Hat USA 2026, researcher Fred Heiding discussed his work on AI-enabled deception and social engineering.
He says AI models are remarkably good at manipulating humans and creating emotional dependency, fueling scams as FBI-reported fraud losses grew from $4 billion in 2020 to about $21 billion in 2025.
Heiding calls the problem asymmetrical: defenders can use AI against technical attacks like vulnerability exploitation, but human psychology can't be patched. He also voices alarm about rogue agent behavior and AI companies' lack of liability.
Paul's Security Weekly covers a record Microsoft patch month, suspected commissary freezer hacks, Defender bypasses, and GPU Rowhammer.
This Paul's Security Weekly news segment reviews Microsoft's September patch release, a record of roughly 972 vulnerabilities including 112 critical, two exploited zero-days, and 20+ potentially wormable bugs, linked to AI-assisted bug discovery.
Hosts discuss simultaneous freezer failures at U.S. military commissaries, where networked refrigeration controls raise questions of a possible cyberattack without proof of one.
Other stories include the defendnot technique that abuses the Windows Security Center API to push Defender into passive mode, Rowhammer attacks on GPUs, mandatory BIOS updates, CVSS limitations, kworker-disguised malware, and a FortiGate post-exploitation RAT.
Security Weekly News marks 25 years since 9/11 and covers AI threats, a Gemini CLI RCE, the Liquid hack return and a Veradigm breach.
This Security Weekly News episode opens by marking 25 years since 9/11 before turning to security news.
Stories include threats hiding in commodity infrastructure, Google's GTIG report on adversarial AI moving toward autonomy, a DEF CON talk on pre-task remote code execution in Google Gemini CLI, and hackers returning $263 million stolen from Liquid Network.
The show notes also cover Anthropic's September 2026 misuse report and a second Veradigm breach, in which stolen vendor credentials were used against a patient-facing API, illustrating third-party credential risk.
Salesforce's Paula Goldman explains how to design AI use cases, decide what agents handle and where humans should lead.
In CXOTalk episode 931, Michael Krigsman speaks with Paula Goldman, Salesforce's Chief Ethical and Humane Use Officer and EVP of Product and author of Manage the Machine.
They discuss how companies choose which work to delegate to AI agents, the measures that show whether a use case works, and what breaks moving from pilot to production.
Topics include how usage-based pricing and tokenomics affect human handoffs, the 'human at the helm' concept, governance for ethical AI, lessons from brands like IKEA, Clorox and Oreo, and the impact on the talent pipeline.
theCUBE's livestreamed summit examined whether contact center AI delivers measurable business value.
theCUBE's AI ROI in Contact Center Summit, held Sept. 11, 2026, featured livestreamed interviews and panels with analysts including Bob Laliberte and Zeus Kerravala and companies such as Five9, Cisco, Talkdesk and Zoom.
It framed contact center AI as shifting from an innovation discussion to an accountability discussion. Sessions focused on whether generative and agentic AI improve cost-to-serve, agent productivity, resolution quality and customer experience, the role of trusted data and knowledge management, and watching recontact rates to avoid false savings.
Event details were verified via SiliconANGLE coverage.
Webinar replay on how a weekly LinkedIn newsletter keeps enterprise sellers present with buyers during quiet deal stretches.
This 31-minute 12touches webinar replay argues that most lost enterprise deals simply drift during quiet stretches, and that follow-up outreach without news reads as nagging.
It explains that a LinkedIn newsletter edition is a LinkedIn Article delivered by email and notification, and that a member account can run up to five newsletters for free.
The session compares newsletters with posts, gives six reasons newsletters work for individual sellers, from advisor positioning to reopening stalled deals, and addresses the challenge of sustaining a weekly publication.
Slides, a companion article round-up, and an FAQ accompany the recording.
Palo Alto Networks' Amit Assaraf explains how shadow AI agents and extensions create endpoint risks that traditional EDR misses.
In Enterprise Security Weekly #475, Amit Assaraf of Palo Alto Networks (former Koi co-founder and CEO) discusses how employee use of local AI models, autonomous agents, and browser extensions is quietly accumulating risk on enterprise endpoints.
He explains why traditional EDR misses non-binary, self-provisioned software. The episode also revisits lessons from the 2023 British Library breach and covers industry news, including Socure's $156M Series E, ClickHouse acquiring RunReveal, a rumor that Proofpoint is in talks to buy Varonis, and research disputing ZeroDayClock's time-to-exploit analysis.
Oxide engineers discuss VOXEL, the emulation environment that lets them test rack configurations they haven't built yet.
In this Oxide and Friends episode, Bryan and Adam are joined by Steve, Andrew, and Rain to discuss how Oxide tests rack configurations it hasn't built yet.
The conversation centers on VOXEL (Virtual OXide Emulation Lab), which runs emulated rack deployments on a single Helios host using Falcon/propolis VMs, SoftNPU software switches, and FRR routers, with optional real service processor firmware.
They explain how emulating components such as the Tofino-based Sidecar switch, Hubris firmware, and trust quorum has accelerated development across many domains of the Oxide stack.
Michael Burtov argues that raising too much money too early can weaken startups by masking problems instead of solving them.
In this video, Michael Burtov explores the idea that capital can become a crutch, replacing the obvious risk of running out of cash with hidden problems like lost urgency, bloated headcount and mistaking growth for product-market fit.
He contrasts heavily funded failures such as Zume Pizza and Quibi with bootstrapped Mailchimp, acquired by Intuit for about $12 billion.
Burtov urges founders to use capital to remove known bottlenecks rather than conceal an unproven business model, discusses how AI may favor capital-efficient companies and revenue per employee, and offers six disciplines for preserving productive constraint.
Tech executive Peter Vardaro shares how transitioning from biomedical equipment repair to IT leadership shaped his philosophy that true developer value lies in solving scalable business problems and saving organizational time rather than just writing code.
In this episode, guest Peter Vardaro speaks with host Mike Kelley about how frustrating software experience as a biomedical technician launched his career in enterprise application leadership.
He emphasizes that IT teams must look beyond isolated support tickets to design scalable systems that remove friction, save thousands of operational hours, and handle HIPAA compliance for protected health data.
Ultimately, Vardaro argues that as AI automates routine coding, tech professionals will find their greatest value in human relationships, deep business context, and critical problem-solving.
AI agents now work with people using enterprise software, but they start out without any knowledge of your organization. Atlassian's product and AI chief explains how to give them context, guardrails, and clear human accountability.
In 2026, AI agents are becoming everyday users of enterprise software built for people. Tamar Yehoshua, Chief Product and AI Officer at Atlassian, explains how to give agents the right organizational context without driving up costs, where human oversight belongs as agents chain together, and how to set measurable goals so AI spending delivers real value.
Key points:
Giving agents the right context, not the most context, drove 44 percent higher quality and 48 percent lower token usage in Atlassian tests.
A named human remains accountable for every agent chain, supported by guardrails, observability, and agents that verify other agents' work.
Broad mandates to simply use AI fall short; teams need specific goals for tenfold process improvements, measured by token spend and value delivered.
Workday Rethinks the Enterprise Workflow for AI Agents
Workday and Google Cloud are collaborating to embed AI agents into existing enterprise workflows, particularly through Google Workspace and Gemini, without requiring users to switch applications.
The partnership focuses on combining probabilistic AI reasoning with deterministic guardrails from ERP systems to ensure compliance with enterprise policies.
Gabe Monroy, Workday's CTO, demonstrated how AI agents can simplify tasks like performance reviews and employee onboarding while maintaining the required approvals and governance controls that traditional applications provide.
Singleton Schreiber CIO Jim Harden explains how IT can shift from break-fix support to business strategy partner.
Jim Harden, a Marine Corps veteran and now CIO at law firm Singleton Schreiber, talks with Phil Howard about moving IT from break-fix to strategy.
He describes giving lawyers one governed entry point for AI workflows using the firm's Microsoft environment, pairing platforms with education, and choosing a platform before models.
Infrastructure as a service cut new-office deployments from days to hours, and growth requires two- and three-year technology plans. Harden also covers preferring focused vendors, a roughly $40,000 project that was scrapped, the value of strong project managers, and leadership lessons.
An IT leader shares how teaching teams replaces becoming the permanent solution.
Demetrios Xarhoulakos discusses transitioning from hands-on technical expert to manager by training support staff instead of personally handling every familiar ticket.
Key themes include breaking down organizational silos between IT, CAD, and development teams through improved communication, and optimizing vendor relationships and licensing through quarterly performance reviews.
He also covers shifting from reactive support toward automation and business enablement. His experience managing global infrastructure migrations, follow-the-sun licensing models, and governed AI adoption shows how domain knowledge strengthens alignment between technology priorities and business strategy.
ServiceNow's chief analytics officer and Axis Communications' CIO discuss when and how to redirect struggling AI projects.
In this InformationWeek podcast, Vijay Kotu (chief analytics officer, ServiceNow) and Jonas Hansson (CIO, Axis Communications) discuss how technology leaders assess and redirect AI initiatives that veer off track.
They address when full project abandonment is warranted versus strategic modification, whether AI projects need different milestones than traditional initiatives, and the risk of intervening too early versus too late.
The conversation includes a tabletop exercise where the guests role-play as interim executives navigating team challenges, acknowledging that early-stage enterprise AI remains largely exploratory.
Cisco and NVIDIA extend their Secure AI Factory partnership to racks drawing more than 200 kilowatts.
The Cisco Secure AI Factory with NVIDIA has been expanded into the rack-scale era, letting enterprises deploy AI across their infrastructure with stronger operational capabilities.
The solution now integrates liquid cooling that supports racks beyond 200 kilowatts for reasoning, agentic AI, and trillion-parameter training.
It pairs NVIDIA Spectrum-X switch silicon with Cisco operating systems and adds integration with Cisco Nexus One management and Cisco Cloud Control for agentic IT operations. Executives frame it as a complete AI factory addressing the gap between buying GPUs and running them in production.
Omdia's Theresa Lanowitz on rising offensive security spend and the risks of agentic AI in red teaming.
Omdia principal analyst Theresa Lanowitz talks with Dark Reading at Black Hat USA 2026 about research showing enterprises increasing investment in penetration testing, vulnerability assessment, and red teaming as attackers weaponize vulnerabilities faster.
Her research found 88% of organizations willing to spend more on offensive security, though they want agents kept under control, integrated with existing tooling, and demonstrably reducing risk.
Concerns about agentic AI include prompt injection, adversarial input, non-deterministic behavior, and token cost - hence her advice to sandbox agents and limit blast radius.
Separate supply chain research found 99% investing but only 44% building an SBOM at build time.
AI observability is evolving from monitoring into an intelligence layer that lets agents diagnose and fix problems.
Paul Nashawaty of theCUBE Research argues traditional observability was built for deterministic software, while AI applications behave unpredictably and require evaluating response quality in context.
Dynatrace's acquisition of Arize AI shows AI and application observability becoming inseparable, since agent failures can stem from models, prompts, APIs, databases or infrastructure.
Observability is shifting from dashboards for engineers to data consumed by autonomous agents that diagnose and remediate issues. Enterprises will require accurate telemetry, shared context, governance and guardrails before granting agents production access.
Mason Herbel of Parallel Systems advises IT leaders to make approved AI tools easy and secure for employees to use.
Mason Herbel discusses moving from an AI startup to Parallel Systems, where governance and company data were the first AI questions.
He suggests enterprise plans, data-handling agreements and trial periods for requested tools, then reviewing actual usage, since blocking sites fails when employees can use personal phones.
He also covers choosing infrastructure workload by workload, including migrating an engineering firm off aging servers and proposing owned GPUs as API bills grew, without claiming verified savings.
Herbel shares how he won support for ideas, watched business workflows first and kept everyday IT covered.
ThreatLocker's Kieran Human explains how sandboxing, least privilege, and monitoring apply to AI agents.
Kieran Human, Lead Cybersecurity Engineer at ThreatLocker, discusses applying established zero trust controls to AI agents on Application Security Weekly.
He describes the properties a good sandbox needs and how monitoring creates a feedback loop that refines allow lists and access controls over time.
The core argument is that an agent's unpredictable behavior is not that different from malware, so the same containment discipline applies.
The segment also covers OWASP's Agentic Skills Top 10, research measuring AI agent exploitation capabilities, and a red agent exploiting a Snowflake vulnerability missed by GitHub Copilot.
More from CxO Podcasts · All IT section newsletters · This week