A day to weaponise, a month to patch
Microsoft published its 2026 Digital Defense Report on Thursday, covering the year to June 2026, and its central finding is a mismatch in timing. The median time from a vulnerability being discovered in the wild to being weaponised has fallen to well below 24 hours. Enterprise remediation of a critical internet-facing vulnerability, the report says, still takes 30 to 60 days.
That is the company’s argument for why it believes attackers currently hold the near-term advantage in the sector’s AI transition. Microsoft’s framing, as reported by Help Net Security, is that “AI is changing the physics of cybersecurity” — compressing the time, information and access an intrusion requires. Work that once needed a specialist, the report argues, in many cases now needs a prompt.
The volume underneath it
The timing claim sits on top of a volume claim. Microsoft counts nearly 40,000 CVEs published in the first half of 2026 alone, putting the year on course to roughly double recent annual totals. Help Net Security reports the company’s full-year estimate at around 72,000.
Neither figure is a measure of AI-written vulnerabilities, and Microsoft does not present it as one. Published CVEs rise for mundane reasons too — more scanning, more coordinated disclosure, more assigning authorities. What the number does establish is the size of the queue that a 30-to-60-day remediation cycle has to work through.
Where intrusions actually start
The report’s intrusion data is less dramatic than its AI framing. Microsoft Defender Experts attributes 30 per cent of initial access to a user executing something they should not have, and another 20 per cent to valid accounts — credentials that worked. Phishing accounted for 23 per cent of investigated intrusions, up from 7 per cent, and exploitation of public-facing applications rose from 15 to 24 per cent.
In credential-based intrusions, Microsoft says 52.2 per cent involved attackers harvesting further credentials once inside, and 18.4 per cent involved password spraying. One campaign Microsoft cites, using the ClickFix technique of persuading a user to paste a supplied command, saw those commands run on more than 1.1 million unique devices between February and early May — roughly eight times the previous level.
The oldest number in the report is the most pointed. CVE-2020-1472, disclosed in 2020, accounted for 58 per cent of detections among the five leading CVEs Microsoft analysed. The gap the report describes is not only about exploits arriving faster. It is also about fixes that were available six years ago and still have not been applied.
How far the autonomy has actually got
Microsoft stops short of saying fully autonomous attacks are the norm. It says they are not. What it reports instead is one controlled evaluation in which a multi-stage autonomous attack ran through 32 stages, and AI-orchestrated activity beginning to appear in the wild rather than only in laboratories.
That is a narrower claim than the headline framing suggests, and worth holding separately from it. A 32-stage chain in an evaluation Microsoft designed is evidence about capability, not about prevalence.
The report lands days after OpenAI said it had notified more than 100 organisations about agents of its own that acted outside their instructions, and after Apple narrowed macOS full-disk access citing risks from AI agents. Three vendors, in a week, describing the same category of problem from three positions.
What to watch is whether the remediation half of Microsoft’s gap moves at all. The weaponisation figure is the one that gets quoted; the 30-to-60-day number is the one a defender controls.
