The One Medical breach claim is what M&A security debt looks like, five years later
Yahoo's sale price got cut by $350 million. Marriott paid £18.4 million in fines. Both breaches traced to systems acquired years earlier, and the same pattern is alleged at Amazon's One Medical.
A breach claim, two old acquisitions, and a five-year gap
In June 2026, the extortion group ShinyHunters claimed to have stolen roughly 8.8 terabytes of data tied to One Medical, the primary care provider Amazon owns. As of this writing, Amazon has not confirmed the scope or validity of the claim, and the figures should be treated as alleged rather than established. What is verifiable is the corporate history underneath the claim, and it traces back further than most coverage has noted.
One Medical itself acquired Iora Health in a deal that closed on 1 September 2021, for roughly $2.1 billion. Eighteen months later, Amazon's acquisition of One Medical closed on 22 February 2023, for $3.9 billion. If the systems named in the breach claim turn out to be the ones in question, the exposure would trace back nearly five years to the Iora deal and almost three years to the Amazon deal, not to anything Amazon built or bought recently.
That gap is the actual story here, more than any single company's specifics. It's a near-perfect illustration of a pattern that has already played out, in public, at least twice before, at very different companies and in very different decades of dealmaking.
It also matters more in a healthcare context than it would somewhere else. The kind of data a primary care provider holds, appointment history, billing details, clinical notes, carries a longer shelf life of usefulness to an attacker than a typical retail breach, and it sits under a stricter set of regulatory obligations. None of that changes the analysis below, but it's part of why a claim against a healthcare acquirer draws more scrutiny, deserved or not, than the same claim would against a generic SaaS company.
Due diligence audits a moment, not a history
Security due diligence in an acquisition is built to answer one question: what does this target's environment look like right now? Acquirers review SOC 2 reports, pen test results, and asset inventories taken at or near signing. That snapshot is genuinely useful for pricing the deal and flagging anything actively broken. What it is not built to do is project forward, past the point where the integration team disbands, the acquired system gets deprioritised against the combined roadmap, and ownership of it quietly becomes nobody's job.
The scale of that gap shows up in the surveys. Deloitte has found that 53% of organisations report a critical cybersecurity issue surfacing after they announce a deal, not before. The Ponemon Institute, in a 2021 study, found that 56% of organisations carry known vulnerabilities that go unresolved for more than a year. Neither number is specific to M&A, but a freshly acquired environment, full of systems nobody at the new parent company fully understands yet, is close to the ideal condition for a vulnerability to sit unresolved for exactly that long.
The mechanism behind that ownership gap is mundane rather than dramatic. A typical acquisition assembles a temporary integration team, gives it a mandate measured in months, and then disbands it once the combined org chart settles. Any system that didn't make the cut for active migration during that window doesn't get a deliberate decision made about it. It just keeps running on whatever credentials, patch cadence, and network access it had on the day the integration team stopped looking, because no one job description includes 'decide what happens to the systems we didn't get to.'
Three deals, three different points of discovery
Lay the three cases side by side and the pattern is less about any one industry and more about how long a breach can sit undiscovered relative to when the deal closed.
Yahoo's case is the fastest-surfacing of the three, and the only one where the breach became public during the deal itself rather than after. Verizon agreed to buy Yahoo's core business in 2016. While that deal was pending, Yahoo disclosed two separate historical breaches: one in September 2016, covering a 2014 intrusion affecting more than 500 million accounts, and a second in December 2016, covering a separate 2013 attack affecting more than a billion accounts. Verizon renegotiated, cutting the price by $350 million to a final $4.48 billion, with Yahoo retaining half the liability for certain post-closing litigation and government investigation costs.
Marriott's case took two years longer to surface, and the deal had already closed by the time it did. The intrusion began in 2014 on a reservation system owned by Starwood Hotels, before Marriott agreed to acquire Starwood in 2016. It went undetected through the entire acquisition and wasn't discovered until September 2018, exposing roughly 383 million guest records. The UK's Information Commissioner's Office initially proposed a fine of £99.2 million in July 2019, then reduced it in stages, first to roughly £28 million, then by a further 20% for mitigating factors such as Marriott's customer notification efforts, and finally down to a confirmed £18.4 million, with the last reduction citing the economic context of the COVID-19 pandemic.
“Yahoo's breach surfaced during the deal. Marriott's surfaced two years after close. The One Medical claim, if confirmed, would put the gap at roughly five years.”
Lay those two confirmed timelines next to the unconfirmed third and a trend line appears even without the third point being settled. The earlier case caught the acquirer mid-negotiation, while there was still room to renegotiate price and reallocate liability. The later case caught the acquirer after the deal had fully closed, with no bargaining position left over the seller at all, just regulators, fines, and reputational cost to absorb alone. If the One Medical claim holds up, it would sit at the far end of that progression: a target two acquisitions removed from the parent footing the bill.
| Deal | When the breach surfaced | Cost |
|---|---|---|
| Yahoo → Verizon (2016) | During the pending acquisition itself | $350M price cut; Yahoo kept 50% of certain post-close liabilities |
| Starwood → Marriott (2016 deal; intrusion began 2014) | 2 years after close (discovered Sept 2018) | £18.4M ICO fine, reduced from a proposed £99.2M |
| Iora Health → One Medical → Amazon (2021 / 2023 deals) | ~3–5 years after close (claimed June 2026) | Unconfirmed as of writing; reputational and regulatory exposure pending |
Why the gap keeps widening instead of closing
A few structural trends point toward longer gaps, not shorter ones. Cloud environments make it easier for an acquired system to keep running quietly without anyone budgeting for its retirement, since there's no physical server prompting someone to ask why it's still there. Deal volume and the pace of add-on acquisitions, buying a company that has itself already bought other companies, mean an acquirer can inherit two or three layers of unowned legacy infrastructure in a single transaction, exactly the shape of the Iora-into-One-Medical-into-Amazon chain.
This isn't a problem confined to technology and hospitality, either. Security researchers tracking M&A-related incidents have found manufacturing accounting for a disproportionate share, around 42% in some datasets, of breaches linked to acquisitions, which suggests the pattern is a feature of how deals get integrated everywhere, not a quirk of any one sector's tech stack.
Chained acquisitions add a multiplier on top of all this. Buying a company that has itself already absorbed two or three smaller acquisitions means inheriting not one integration backlog but several, stacked at different ages and documented to different standards, if they're documented at all. The parent company at the end of that chain usually has the least visibility into the oldest layer, precisely because it's furthest removed from whoever originally built or bought it.
What due diligence would have to do differently
None of this argues against doing acquisitions. It argues against treating the security review as something that finishes at signing. A handful of changes would close most of the gap:
- Inventory systems by origin, not just current network presence, and flag anything that arrived through a prior acquisition the target itself made.
- Assign an explicit, named owner to every inherited system before close, not as a follow-up task after integration starts.
- Set a default decommission-or-modernise deadline for legacy systems, so 'it still works' stops being a permanent justification for leaving it alone.
- Re-run security assessments on acquired infrastructure on a fixed schedule after close, not only once at signing.
- Treat a target's own acquisition history as part of the attack surface being purchased, and diligence it accordingly.
- Track the integration backlog as a security metric, not only a project-management one, since backlog age is a reasonable proxy for unowned risk.
None of these are exotic recommendations, and that's the point. They're closer to basic asset management than to advanced security engineering. What's missing in most deals isn't the technical capability to do them. It's a named owner with a mandate to keep doing them after the team that cared most about the acquisition has already moved on to the next one.
The lesson isn't to avoid deals. It's to keep auditing after they close
Yahoo, Marriott, and now, pending confirmation, One Medical didn't get hit because their acquirers skipped due diligence. They got hit because due diligence is built to answer a question about the day of signing, and the risk that actually materialised lived in the years afterward, in systems nobody had explicitly agreed to keep watching. The fix isn't a better one-time audit. It's accepting that the audit clock shouldn't stop when the deal does.
Frequently asked questions
Related reading
Microsoft 365 Went Down With No Traffic Spike. Load Balancer Health Checks Need a Removal Limit.
Microsoft said load rose because servers were removed, not because traffic grew. Here is how health checks turn into a cascade, and the removal floor each major load balancer offers.
Zero Data Retention Covers a Request Path, Not a Vendor. The Fireworks Breach Reached the Paths Outside It.
The Fireworks AI breach exposed Notion AI transcripts despite a zero data retention commitment. The promise applied to inference requests. The transcripts sat in a feature that stores by design.
Daylight Saving Ends 25 October (EU) and 1 November (US). Know What Your 01:30 Scheduled Job Does
Two clock changes arrive in the next two weeks. Here is what daylight saving does to cron jobs, application schedulers and "daily" reports, and how to decide which jobs belong in UTC.