Key takeaways
- HFS Research and Rackspace Find Only 33% of Enterprises Have Met the Business Case for Their Biggest AI Program, While 96% Report Governance Gaps. Treat governance as the critical path rather than the cleanup task. Before approving the next AI use case, require a named accountable owner, one consistent production operating model, and a business case that distinguishes efficiency gains from revenue claims. If your portfolio is spread across many functions, consider consolidating around the few places where AI is already embedded in standard operations, because that is where this study found the strongest delivery record.
- IBM and Microsoft Launch Imagine Enterprise With More Than 6,000 Prebuilt AI Agents and Assets Through a Services Led Delivery Model. Ask three questions before you evaluate Imagine Enterprise: what the pricing model is, which agent protocols it supports for interoperability, and what exactly counts inside the 6,000 agents and assets. Then decide whether a services led, vendor operated model fits your control requirements, or whether the self management path gives you the independence you need. Do not let a large library of prebuilt assets substitute for your own architecture decisions.
- The Justice Department Creates a Super Intelligence Task Force That Can Ask Prosecutors to Notify It Before They Open a Case or Seek Charges. Do not wait for the task force to define expectations. Build an inventory of where AI operates in your business, including tools bought from vendors, assign a named owner for each material AI risk, and document the controls and human oversight around high risk uses. If a federal inquiry ever touches your AI systems, a current, written record of those decisions will be your best evidence of a working compliance program.
- Arctic Wolf Reports That AntiFox, an Open Source Agentic Pentesting Platform Hosted on Alibaba Cloud, Was Used to Steal AWS Credentials. Assume that attackers now have agents that enumerate cloud environments at machine speed. Prioritize detection on secret access from unfamiliar infrastructure, remove credentials from configuration files and rotate any that were exposed, tighten metadata service settings, and make sure management and data event logging is on before an incident rather than after one. Measure your detection time against minutes, not hours.
Welcome to the DX Today Executive Briefing
This week the evidence about enterprise AI split into two stories that executives should read together. On one side sit the buyers, who in a new HFS Research and Rackspace study say that only 33% of enterprises have met the business case for their most significant AI program. On the other side sit the vendors, the regulators and the attackers, all of whom are moving faster than most governance programs.
In this edition we cover that study, the launch of IBM and Microsoft Imagine Enterprise with more than 6,000 prebuilt agents and assets, a Department of Justice memorandum that created a Super Intelligence Task Force, and an Arctic Wolf report on AntiFox, an agentic pentesting platform used to steal AWS credentials. Each story comes with a Strategic Takeaway for the executive who has to act on it.
01
HFS Research and Rackspace Find Only 33% of Enterprises Have Met the Business Case for Their Biggest AI Program, While 96% Report Governance Gaps
A study released on November 12, 2026 by HFS Research and Rackspace puts a hard number on a feeling that many boardrooms already have. According to the release, only 33% of enterprises have met or exceeded the business case for their most significant AI program, while 96% report significant governance gaps. The result comes from a survey of 502 senior enterprise AI decision makers, fielded between August and September 2026 in partnership with Rackspace.
The sample is worth understanding before anyone quotes the headline. The release says respondents worked at enterprises with at least $100 million in annual revenue across eight industries and four regions, and that 83% were already operating AI in production or beyond. In other words, this is not a survey of curious laggards. These are organizations that have moved past pilots, which makes the 33% figure more sobering rather than less.
The study locates the problem in governance rather than in technology. According to the release, 78% of respondents view governance failures as a bigger threat to scaling AI than technology failures, and 44% of enterprises have no single, consistent approach to running AI in production. That second number describes an operating model problem. When nearly half of the organizations running AI in production lack one consistent way of running it, every new use case becomes a bespoke project with its own owners, controls and failure modes.
The most counterintuitive finding concerns scale. The release reports that 37% of enterprises deploying AI in one function have met or exceeded their business case, compared with 28% of those deploying across several functions. Breadth, in other words, made delivery harder rather than easier. The same release reports that 56% of organizations with AI embedded in standard operations have met or exceeded their business case, which suggests that value arrives when AI stops being a project and becomes part of how work is normally done.
The gains that executives are actually seeing are also uneven. In the study, 58% of respondents cite workforce productivity improvements and 42% cite measurable cost reductions, compared with 28% reporting better customer experience and 21% reporting measurable revenue growth. Efficiency is paying out first, while growth and customer outcomes lag well behind. A board that approved AI spending on the promise of new revenue should read that last number carefully.
The release then makes its commercial argument. It reports that 61% of enterprises prefer one accountable partner across AI models, platforms and infrastructure, rising to 74% among production stage firms whose programs have not delivered, and that 73% of respondents say they would pay a premium over raw infrastructure for a fully managed AI service with governance accountability. It also reports that 63% of respondents recognize the proposed operator role as meaningfully different from what they buy today. Phil Fersht, Founder and CEO of HFS Research, and Gajen Kandiah, CEO of Rackspace, are both quoted in support of that case.
Readers should weigh the framing as well as the data. The study was conducted in partnership with Rackspace, and its conclusion points toward the kind of accountable, fully managed operator model that the release itself proposes. The release states that the survey did not name any vendor, platform or provider, and that its findings describe AI active enterprises, not market wide AI adoption. The report is titled Run Your AI as One, or Keep Paying for Parts That Never Add Up. The numbers are useful as a benchmark, but they are a sponsored benchmark, and they should be read as one input rather than as a verdict.
Strategic Takeaway
CIOs, CFOs and Enterprise AI Program Owners
Treat governance as the critical path rather than the cleanup task. Before approving the next AI use case, require a named accountable owner, one consistent production operating model, and a business case that distinguishes efficiency gains from revenue claims. If your portfolio is spread across many functions, consider consolidating around the few places where AI is already embedded in standard operations, because that is where this study found the strongest delivery record.
02
IBM and Microsoft Launch Imagine Enterprise With More Than 6,000 Prebuilt AI Agents and Assets Through a Services Led Delivery Model
On November 10, 2026, IBM and Microsoft announced Imagine Enterprise, which the release describes as an agentic platform designed to help enterprises build, run and govern AI agents. The pitch is that large companies should not have to assemble agent infrastructure from parts. They can start from a platform that arrives with prebuilt work already inside it.
The headline number is scale. According to IBM, IBM Consulting brings more than 6,000 prebuilt AI agents and assets built on IBM Consulting Advantage to the platform. It is worth reading the phrase carefully, because the release counts agents and assets together, so the figure should not be read as 6,000 finished, independent agents. It is better understood as a library of starting points that clients can adapt.
The platform combines technology from both companies. The release says Imagine Enterprise draws on Microsoft Foundry, GitHub Copilot and Microsoft Copilot Studio, along with IBM Consulting Advantage components named Process Studio, Context Studio and Advantage Core. For buyers, the practical meaning is that the platform sits at the intersection of two large enterprise estates, and it will be most natural for organizations that already run on both.
The launch is organized by industry rather than by technology. Imagine Enterprise starts with initial industry frameworks for financial services, insurance and retail, with manufacturing, energy, public sector and mobility named as future candidates. SEI is the featured customer in the announcement. The release names Mohamad Ali, Senior Vice President and Head of IBM Consulting, Rodrigo Kede Lima, President of the Microsoft Frontier Company, and Sneha Shah, Chief AI Strategist and Head of SEI Next at SEI.
Delivery is the part of the announcement that deserves the most scrutiny. IBM says the platform is available to clients by engaging IBM Consulting to deploy and operate it on the client's behalf, or by co innovating with IBM Consulting, and adds that self management options are also available. The release also offers forward deployed engineering support for qualifying engagements, without stating the qualification criteria. This is a services led launch. Buyers should expect that getting value from Imagine Enterprise will involve a consulting relationship, whatever the self management options turn out to be.
Several questions remain open for buyers, and they are the ones to put to both vendors before signing anything. Executives should ask what the pricing model is, when the platform will be generally available for their industry, and how agents built on the platform will interoperate with agents built elsewhere, including on open protocols such as MCP and A2A. A platform that governs agents should be able to explain how those agents work with everyone else's, and a services led launch makes it especially important to get those answers in writing.
There is also a distribution angle. AI Weekly reported that fewer than 7% of Microsoft's 450 million Microsoft 365 seats carry paid Copilot licenses, and presented that gap as the reason a joint offer aimed at regulated industries could matter. That figure is the publication's reporting and not part of the IBM release, so treat it as context rather than as a company statement. Even so, it frames the strategic logic: Microsoft's installed base still has far more room to grow into paid AI than it has captured so far.
Strategic Takeaway
CTOs, Chief Architects and Heads of Enterprise Platforms
Ask three questions before you evaluate Imagine Enterprise: what the pricing model is, which agent protocols it supports for interoperability, and what exactly counts inside the 6,000 agents and assets. Then decide whether a services led, vendor operated model fits your control requirements, or whether the self management path gives you the independence you need. Do not let a large library of prebuilt assets substitute for your own architecture decisions.
03
The Justice Department Creates a Super Intelligence Task Force That Can Ask Prosecutors to Notify It Before They Open a Case or Seek Charges
A memorandum dated October 13, 2026 established the Department of Justice Super Intelligence Task Force, and its implications are now reaching corporate legal teams. The memorandum cites Executive Order 14434 of September 29, 2026, titled Inaugurating the Era of Super Intelligence. PYMNTS described the memorandum as coming from Attorney General Todd Blanche. The practical effect is a single coordinating body for how the department handles matters that involve super intelligence.
The mandate has two halves. According to the memorandum, the task force will ensure that America's leadership in super intelligence is not compromised by the misuse of super intelligence, and will take affirmative measures to promote American super intelligence dominance. One half is enforcement. The other is promotion. That combination signals that the department sees itself as both a policeman and a supporter of the sector, which is a different posture from a pure enforcement unit.
The structure is senior. The memorandum says the task force is chaired jointly by the Attorney General or designees Associate Attorney General Stanley Woodward and Associate Deputy Attorney General Stephen Weiss. Members include representatives of the Antitrust Division, Civil Division, Criminal Division, FBI, National Fraud Enforcement Division, National Security Division and the Office of the Solicitor General. A body that spans antitrust, fraud, national security and the courts can look at an AI matter from every angle at once.
The most operationally important sentence is about notification. Under the memorandum, components and United States Attorneys' Offices should, from time to time at the request of the task force, notify it upon seeking to open an investigation and before seeking charges or finalizing a resolution in matters involving super intelligence. The memorandum says this should not delay action and is not necessary in exigent circumstances. In practice, that means AI related cases are likely to be seen by a central group before they are brought, which should make enforcement more consistent and probably more deliberate.
The task force also consolidates earlier work. The memorandum merges the Artificial Intelligence Task Force, announced on January 9, 2026, into the new task force, so that the earlier group ceases to operate as a separate body. The memorandum is also careful about its own legal force. It states that it does not create any right or benefit, substantive or procedural, enforceable at law or in equity. It is an internal management instrument, not a new rule for companies.
For businesses, the signal is in the design rather than in any new rule. A task force that sees matters before they are opened and again before charges or resolutions are finalized is positioned to build a consistent view of how AI cases should be handled. That consistency cuts both ways. It should reduce surprise for companies that run credible programs, and it raises the cost of improvisation for companies that cannot explain how their AI systems are governed.
In our analysis, the sensible response is to prepare for the questions a prosecutor would ask, which are what a company knew, what it controlled and who was accountable. AI adds new answers to each of those questions: which systems are in use, including tools bought from vendors, who owns each risk, and what human oversight exists over high risk uses. The memorandum does not say what compliance programs must contain, so this is preparation rather than obligation. Companies that can answer in writing will be in a stronger position than companies that cannot.
Strategic Takeaway
General Counsels and Chief Compliance Officers
Do not wait for the task force to define expectations. Build an inventory of where AI operates in your business, including tools bought from vendors, assign a named owner for each material AI risk, and document the controls and human oversight around high risk uses. If a federal inquiry ever touches your AI systems, a current, written record of those decisions will be your best evidence of a working compliance program.
04
Arctic Wolf Reports That AntiFox, an Open Source Agentic Pentesting Platform Hosted on Alibaba Cloud, Was Used to Steal AWS Credentials
Arctic Wolf Labs Adversary Research published a report dated November 10, 2026 titled Antifox Agentic Pentesting Platform Used to Steal AWS Credentials and Move Laterally in Compromised Environments. BleepingComputer covered it on November 11, 2026, reporting that Arctic Wolf researchers found hackers using AntiFox AI agents in AWS credential theft attacks, and Security Affairs reported the same day that Arctic Wolf Labs published research on attackers using AntiFox. The story matters because it shows a tool built for defenders being used by attackers.
Arctic Wolf describes AntiFox as an open source, Chinese language agentic pentesting platform. The research says that Alibaba Cloud systems hosting AntiFox interfaces were associated with confirmed malicious actions against internet facing web services and associated AWS environments. In plain terms, attackers appear to be running an AI agent that automates penetration testing, from servers on one cloud, against the customers of another.
BleepingComputer added lineage. It reported that AntiFox was initially created as a fork of ARTEX 0.1.1 and has been updated to version 2.0, and that ARTEX was used in a CrowdStrike linked Vapor Panda campaign that targeted more than 200 organizations across 14 countries. Readers of this briefing will remember ARTEX from earlier coverage of that campaign. The pattern is that open source agentic frameworks are being forked and reused, so defenders face a family of tools rather than a single threat.
The speed of the activity is the first lesson. In one intrusion, Arctic Wolf says the AntiFox linked source cycled through multiple AWS identities and generated hundreds of cross service API requests within approximately half an hour. A human operator would struggle to make that many deliberate requests so quickly. An agent does it as a matter of routine, which compresses the window defenders have to notice and respond.
The entry points were ordinary. Arctic Wolf says initial access varied by intrusion, with either a compromised internet facing IIS application or a valid AWS credential obtained through unknown means, and that the exact vulnerability used for code execution was not identified. Arctic Wolf observed an ASPX web shell created by the IIS worker process, commands that read web.config files and connection strings, and in some cases a successful GetSecretValue call against a production secret. That chain, from a web server to a configuration file to a cloud secret, is a reminder that application hygiene and cloud identity are one problem.
Amazon's position should be recorded alongside the research. An Amazon spokesperson said AWS services were not compromised and operated as designed. That statement is consistent with the research, which describes compromised web applications and credential misuse rather than a flaw in AWS itself. The responsibility sits with how customers store, scope and monitor their credentials.
The defensive guidance is specific. Arctic Wolf recommends enabling CloudTrail management events and S3 data events for sensitive buckets, alerting on successful GetSecretValue calls from novel infrastructure, rotating credentials exposed through web.config files, and enforcing IMDSv2 with a restrictive metadata response hop limit where EC2 instance roles are required. None of these steps is exotic, and all of them reduce the value of a stolen credential to an automated attacker.
Strategic Takeaway
CISOs, Cloud Security Leaders and Heads of Security Operations
Assume that attackers now have agents that enumerate cloud environments at machine speed. Prioritize detection on secret access from unfamiliar infrastructure, remove credentials from configuration files and rotate any that were exposed, tighten metadata service settings, and make sure management and data event logging is on before an incident rather than after one. Measure your detection time against minutes, not hours.
The Analysis
The Bottom Line
Taken together, this week's stories describe an enterprise AI market that is racing ahead of its own controls. The HFS Research and Rackspace study says only 33% of enterprises have met the business case for their most significant AI program and 96% report significant governance gaps. IBM and Microsoft are answering with a services led platform of more than 6,000 prebuilt agents and assets, while the Justice Department is building a central task force and Arctic Wolf is documenting attackers who use agents of their own.
The common thread is accountability. Buyers say they want one accountable partner, prosecutors will ask who was accountable for AI risk, and defenders must show accountability for every credential an agent can reach. Technology choices matter, but the organizations that pull ahead will be the ones that can name an owner, a control and a record for each AI system they run.
Our advice for the week ahead is practical. Inventory your AI, assign owners, ask your platform vendors about pricing and protocols before you commit, and test your cloud detection against the speed of an agent. The gap between the 33% and everyone else will be closed by operating discipline rather than by another announcement.