The methods have not changed and the targets have not changed either, yet what has shifted decisively this year is the blast radius of any single compromise.

That is the uncomfortable thread running through the 2026 Microsoft Digital Defense Report, which draws on the trillions of security signals Microsoft processes every day. Attackers are still pursuing the same prizes they have always wanted, namely identities, credentials, and data, but the systems we have built to run a modern business, among them identities, AI services, cloud platforms, software supply chains, edge devices, and critical infrastructure, have become so tightly interconnected that a single compromise no longer stays where it lands, and instead it ripples outward, because one stolen credential, one poisoned software package, or one unpatched edge appliance can carry its effects through every connection that the affected organisation once trusted.

For Indian enterprises that are building digital public infrastructure and racing to adopt AI at national scale, that shift is anything but abstract, because the very connections that make our economy fast are the same connections an attacker now uses to turn one breach into everyone’s problem.

Trust itself is now the attack surface

For two decades we secured the perimeter first and then we secured identity, yet the report’s central finding is that attackers have climbed another level again, because they are now exploiting trust itself, meaning the trusted relationships between people, services, suppliers, and platforms that allow interconnected systems to function at all.

The numbers make the point more sharply than any framework can, since user execution, valid accounts, social engineering, and phishing together account for 73.3 percent of initial access attempts, which tells us that attackers are not so much breaking the door down as being handed the keys. Somewhere between 89 and 95 percent of email phishing attachments led to a credential theft effort, more than 46 million business email impersonation attacks were detected, and 93 percent of voice phishing attempts kept the victim on the line long enough to begin social engineering, which is a sobering reminder that the human being remains the most reliable exploit in the catalogue.Once the attacker is inside, the pattern is depressingly consistent, because more than half of intrusions lead to follow-on credential theft, after which the intruder harvests further identities, sprays them internally, and watches one compromised account quietly become many. Microsoft Incident Response observes the same chain again and again, a chain that begins with the compromise of a human identity, moves to the discovery of a non-human one such as a service principal or an API key, escalates privilege using access that was already present, and ends in actions on objectives, with 63 percent of intrusions culminating in data theft. None of this amounts to exotic tradecraft, which is precisely the point, because it is ordinary trusted machinery being turned against the very people who rely on it.

AI moved the clock, and the clock is not on our side

What was once a threat on the horizon has now arrived, and AI is changing the economics of attack and defence in a way that hands the near-term advantage to the attacker. For the more sophisticated actors, AI compresses the attack chain from days into seconds, whether the task at hand is finding vulnerabilities, generating custom malware, shortening the cycle of lateral movement and exfiltration, or orchestrating entire attack chains, while for the less sophisticated it makes nation-state-grade persistence and customised social engineering available more or less off the shelf.

The tempo of all this shows up everywhere you care to look, since exposed cloud workloads were attacked within an average of just 5.3 hours, and the window between a vulnerability being disclosed and being exploited has narrowed to a matter of days, a gap that AI-accelerated scanning is compressing further still. State-aligned operations out of China, Iran, North Korea, and Russia are meanwhile converging on the same quiet playbook, which is to embed inside legitimate authentication flows, collect patiently from the cloud, and pre-position inside high-value environments for the long term.

Here is the part that leaders too often skip, because if attacks now arrive at machine speed then they can only be met at machine speed, which makes AI a requirement for defenders rather than a luxury and means trusting agentic systems with genuine autonomy to investigate and, within defined limits, to act. That reality raises a governance question India’s boards can no longer defer, since every defensive agent you deploy is itself a new non-human identity that carries standing access across your estate. The report is explicit that organisations must now govern human and non-human identities together, maintain visibility into their agentic capabilities, and report agent permissions as a board-level risk metric, because an agent that you cannot attribute, scope, or revoke inside the breakout window is not really a control at all, but rather exposure wearing your own badge.

The India weight: accountability does not thin out because software did the work

Two forces make this especially pointed in the Indian context. The first is that India’s digital economy runs on shared, interconnected public and private infrastructure, which is exactly the kind of dependency the report warns will cascade fastest, and the second is that under India’s Digital Personal Data Protection Act of 2023, accountability sits squarely with the organisation acting as data fiduciary. That obligation does not dilute simply because an AI system or an agent processed the data rather than an employee, and so the familiar excuse that the model did it will never serve as a defence. In an interconnected and increasingly automated estate, attribution, by which I mean knowing which identity, whether human or machine, did what and on whose authority, becomes a compliance capability rather than merely a forensic one.

From defence-in-depth to resilience by design

For the better part of a decade we optimised against named threats, building one defence for ransomware and another for distributed denial of service, yet the report argues convincingly that the era of isolated threats is now behind us. We operate instead in a polycrisis, in which geopolitical pressure, economic stress, rapid technological change, and relentless cyber activity overlap with almost no breathing room in which to recover, and defences that were tuned for one threat at a time tend to fail precisely when several crises arrive together.

The redefinition at the heart of this year’s report is simple to state and demanding to live by, because security is no longer about protecting individual systems so much as it is about understanding interconnected risk and strengthening resilience across the whole ecosystem. Where defence-in-depth assumed that you could keep the attacker out, resilience by design assumes instead that failure will happen, treats dependencies as seriously as it treats adversaries, and makes recovery and continuity every bit as important as prevention. The defining question for an Indian board in 2026 is therefore no longer whether disruption will occur, but whether the organisation can keep operating, contain the harm, recover quickly, and hold on to trust when it does.

What to actually do: ten priorities that survive contact with AI

The report wisely resists the urge to publish yet another framework, and it offers instead ten priorities that remain durable as AI multiplies the identities, applications, and agents an organisation has to govern, all of them resting on secure foundations, namely Secure by Design, Secure by Default, and Secure Operations, applied through an identity-centric model of Zero Trust.

Two of these priorities carry extra weight in the Indian context. Collective defence amounts to far more than a slogan, because the report’s own data shows that the correlation of signals within and between organisations is the single strategic variable that defenders most directly control, and every additional source of visibility strengthens the value of every other. For a market that is still maturing its national coordination, a streamlined report-once mechanism, together with security operations centres that are wired into sector-level structures, would move the needle further than any single product could, while the disciplined governance of AI and non-human identities is the control that stops this year’s productivity story from becoming next year’s incident report.

The honest close

Trust between human beings is something we infer from history, intent, and reputation, yet none of those signals survives the move to machine actors and interconnected systems, where authority now travels further and faster than anyone ever intended. So we stop inferring and start instrumenting, which in practice means giving every actor, whether human or machine, a scoped identity, checking authority at the precise moment of action, ensuring revocability inside the breakout window, and producing evidence that a board can actually read without needing a translator.

Zero Trust taught us to stop assuming the network, and the interconnected, AI-accelerated enterprise now asks something considerably harder of us, which is to engineer for endurance on the working assumption that a compromise somewhere in the web of dependencies we rely on is not a distant possibility but an ordinary Tuesday. The organisations that will lead in this environment are not the ones that somehow never get hit, but rather the ones that can still operate, and can still be trusted, on the days when they are.

Note: This article is a part of ETCISO’s Brand connect Initiative.

Join the community of 2M+ industry professionals.

Subscribe to Newsletter to get latest insights & analysis in your inbox.

All about ETCISO industry right on your smartphone!