All Quiet on the New Frontier

Or, why too quiet is the most dangerous sound in security.

All Quiet on the New Frontier. Or, why too quiet is the most dangerous sound in security. Red points of light scattered over a dark, dotted landscape, one of them flaring. 12 of 12 unknown flaws found by one AI system.
Studio 111 Labs · 13 November 2026 · 7 min read · Figures as published at time of writing

In January, an AI read the code behind the padlock on much of the internet and found twelve flaws that people had missed. Some had been sitting there since 1998. No alarm rang.

01The lock

You have used OpenSSL today. You have never opened it, paid for it or agreed to it.

It is free software that scrambles what travels between two computers so nobody in the middle can read it. When your browser shows a padlock, a large share of the time OpenSSL is the lock on the other end. It is maintained by a foundation and it has been read, tested and attacked by experts for more than twenty-five years.

On 27 January 2026, OpenSSL shipped a security release that fixed 12 previously unknown flaws, the kind the industry calls zero-days. All twelve were found by one AI system, built by a company called AISLE. The foundation's chief technology officer put it plainly: “This release is fixing 12 security issues, all disclosed to us by AISLE.”

AISLE says some of them dated back to 1998. Two affect code going back to OpenSSL 1.0.2, an older generation of the software that many systems still run.

12/12
Flaws in the release found by a single AI system
1998
How far back some of them went, according to the company that found them
0
Attacks reported using them when the fix shipped. Defenders got there first

The twelve

Every flaw fixed on 27 January 2026, in plain English, at the severity OpenSSL gave it.

  1. HighOpening a booby-trapped encrypted message could let an attacker run their own codeCVE-2025-15467
  2. ModerateMemory overflow while checking a certificate fileCVE-2025-11187
  3. LowCrash in a newer, faster kind of web connectionCVE-2025-15468
  4. LowData silently cut short in next-generation signaturesCVE-2025-15469
  5. LowA connection that can eat a server's memoryCVE-2025-66199
  6. LowMemory corruption in how text is bufferedCVE-2025-68160
  7. LowAn encryption mode that did not fully encrypt on some chipsCVE-2025-69418
  8. LowMemory corruption reading names in a certificate fileCVE-2025-69419
  9. LowCrash while verifying a timestampCVE-2025-69420
  10. LowCrash while decrypting a certificate fileCVE-2025-69421
  11. LowCrash while reading a certificate fileCVE-2026-22795
  12. LowCrash while verifying a digital signatureCVE-2026-22796

Source: AISLE's disclosure and OpenSSL's advisory, as summarized by SecurityWeek and Datadog Security Labs. Descriptions are our plain-English paraphrase.

Now the part a scarier article would skip. Ten of the twelve were rated low. Datadog's security team, who tested them, said a crash was the most realistic outcome and that ordinary website traffic was not the exposed path. Nobody has reported an attack that used them. They were found by people who wanted them fixed, and they were fixed.

This is the good ending. It is still the most unsettling security story of the year, for one reason.

The flaws were always there. For years, in some cases decades, the correct status report on every one of them was the same two words. Nothing happened.

02Your day runs on it

When Netcraft measured it in 2014, the two web servers that use OpenSSL by default, Apache and nginx, sat behind two-thirds of the active websites in the world. It is also bundled inside developer tools such as Node.js, and inside mail gateways and the programs that handle certificates.

It is not everything. Some software carries its own lock. But try getting through a day without touching it.

One ordinary Tuesday

Each row is an encrypted connection, and the red line is what rides on it.

  1. 06:45HOME
    Your phone syncs mail and messages
    Your password, your contacts
  2. 07:30HOME
    You check a balance in the bank app
    Account numbers, your login
  3. 08:10HOME
    You tap a card for coffee
    Card details, from the terminal to the processor
  4. 09:00WORK
    You sign in to email and the VPN
    The keys to everything the company owns
  5. 10:30WORK
    A customer pays on your checkout page
    Their card, and your name on the receipt
  6. 13:15WORK
    You upload a signed contract to cloud storage
    A client's terms, pricing and signatures
  7. 15:40WORK
    Someone pastes a spreadsheet into an AI tool
    Whatever was in it. Hold that thought
  8. 18:20HOME
    You refill a prescription online
    Health records, date of birth, insurance
  9. 21:00HOME
    The doorbell camera uploads its clips
    Who comes and goes from your house

Illustrative day, not data. OpenSSL is one of the most common pieces of software making connections like these on the server end; we have not checked which library any particular service runs.

We already know what a bad day on this lock looks like. On New Year's Eve 2011, a small piece of code was added to OpenSSL. It shipped in 2012. In April 2014 the world learned it was broken. Netcraft found the flaw affected around 17% of trusted secure web servers, about half a million of them.

It was called Heartbleed. It sat in the open for more than two years, in some of the most scrutinized security code there is, and for more than two years the status report was: nothing happened.

03You only learn from what you feel

People are not bad at risk. Touch a hot pan once and you never need the lesson again. Step too close to a ledge and your body handles it before you think.

We are bad at one specific kind of risk: the kind where the benefit is immediate and certain, and the cost is delayed, invisible, and impossible to trace back to the thing you did.

Security lives almost entirely in that second kind. IBM studies hundreds of breached organizations every year. In its 2026 report, the average breach took 247 days to identify and contain. For the first 183 of those days, nobody inside the company knew.

247days, on average, to identify and contain a breach. One square is one day. These are companies with security teams, tooling and alarms.
183 days: breached, and nobody knows64 days: known, not yet contained

IBM Cost of a Data Breach Report 2026: mean time to identify (183 days) and contain (64 days).

Eight months. And that is the short one.

How long the silence lasted

Same scale, top to bottom. The time between a problem existing and anyone feeling it.

The oldest of the twelve
1998 → January 2026
Heartbleed
December 2011 → April 2014
The average breach
247 days
A prompt pasted into a personal AI account
No alarm exists, so the clock never starts
No number to draw

Dates from AISLE, Netcraft and IBM. Bar lengths are our arithmetic on the dates shown. The last row is not data. It is the absence of any.

04Two threats. You are only picturing one.

The swarm. The system that found the twelve was pointed at OpenSSL by people who wanted it fixed. Nothing about the technique requires good intentions.

Picture ten thousand AI agents working every hour of every day, testing every door of every company at once. The ten thousand is our picture, not a count. The capability behind it is real. In August, Palo Alto's Unit 42 said its autonomous platform had found more than 14,000 previously unknown vulnerabilities across nearly 4,000 open-source projects in two months.

Google's Mandiant reports that attackers spent 2026 moving from asking chatbots questions to running autonomous attacks, and that exploiting a vulnerability has been the most common way in for six years running.

The office. Now go back to 15:40 on that Tuesday. Picture a hundred employees. Nobody is attacking them. They are busy, the tool is good, and the answer comes back in ten seconds.

Browser telemetry from LayerX found that 77% of employees who use AI tools paste data into them, and 82% of that pasting goes through personal accounts the company cannot see. Cyberhaven's 2026 report found that 39.7% of AI interactions involve sensitive data, and that the average employee puts sensitive data into an AI tool once every three days.

By our arithmetic, at that rate a 100-person company sends something sensitive into an AI tool about 33 times a day. Call it 12,000 times a year. Both firms sell monitoring software, so discount the figures as you like. Cut them in half and it is still a firehose.

One of these has an alarm industry. The other has a paste button.

Left: 10,000 dots, one per imagined agent. An illustration of scale, not a measurement. Right: 100 employees who use AI tools, with the share who paste data into them lit.

The swarm

10,000

Ten thousand dots, one per imagined AI agent.
  • Comes from: outside, uninvited
  • Works: every hour, every door
  • You find out: late. 247 days on average
Loud, eventually

The office

77 of 100

  • Comes from: inside, voluntarily
  • Works: between meetings, on a personal login
  • You find out: possibly never
Silent

Right panel: LayerX Enterprise AI and SaaS Data Security Report 2025, browser telemetry; the denominator is employees who already use generative AI tools, not all employees. Left panel is schematic.

The swarm is the one that keeps people up at night, and it should. It is also the one with firewalls, scanners, insurers and an entire profession pointed at it. When OpenSSL has twelve flaws, somebody ships a fix.

The office has none of that. Nothing breaks. No invoice arrives. Nobody ships a fix for what your team typed last Tuesday.

Until you see the impact, you assume it's fine.

05Nothing happened

Ask anyone who has pasted a P&L, a client contract or a pricing sheet into a chatbot whether it was a mistake. You will get the same two words the lock gave us for decades. Nothing happened.

Translate that sentence. It means: nothing I could see happened, in the time I have been looking, through the channels where I would have heard about it.

We want to be straight about the limits here. Nobody has shown that one company's prompts turn up in a competitor's answers, and this piece does not claim it.

But ask what the evidence would look like if it had cost you something. A bid you lost by a hair. A competitor who seemed to know your margins. A setting that changed in a terms update you did not read. None of those arrive with a label saying which Tuesday afternoon paste they came from.

The same six months, from two chairs

Illustrative. The log on the right is invented to show the shape of the problem, not taken from any company.

What you saw
0

Alerts. Complaints. Invoices. Calls from a lawyer. Reasons to think about it at all.

What left the building
  1. Mon 09:14Q3 P&L, pasted to "clean up the formatting"
  2. Mon 14:02Client contract, pasted to "summarize the terms"
  3. Tue 15:40Customer list with emails, uploaded to "dedupe"
  4. Wed 16:27Pricing sheet and margins, pasted to "build a quote"
  5. Thu 11:05Source code with an API key in it
  6. Fri 08:51A termination letter, with the employee's name

Synthetic example. Every item on the right went through a personal account, so no system the company owns recorded it.

OpenSSL got lucky in the best way. Someone built a tool, pointed it at the lock and counted the flaws. You are running your own experiment in private, and nobody is counting.

06Build the loop yourself

Telling people to care more will not work. Caring was never the bottleneck. People respond to what they can see, so the job is to make the exposure visible before the cost does it for you.

Three questions will tell you where you stand.

Question one

OpenSSL shipped its fix on 27 January. When did yours arrive? Name the person who updates the servers, the website plugins and the office devices. If nobody knows, that is the finding.

Question two

If sensitive data had been leaving for six months, what would you have seen? Name the report, the alert or the person. If you cannot, you have your answer.

Question three

Which login is each person actually using? Not the plan the company bought. The account open in the browser right now.

If the honest answer to the second question is "nothing," then nothing happened is all you will ever hear. The lock said the same thing for decades.

Studio 111 Labs · AI Exposure Review

Find the gaps before someone else does.

We help small and mid-sized businesses close the loop. It starts with a short review of the AI tools, logins and terms already in use, and ends with a one-page map of what is going where and what should change. No hardware pitch and no vendor to sell you. If you would rather talk it through first, that is how most of our work starts.

Start a conversation →
Sources
AISLE, Discovered 12 out of 12 OpenSSL Vulnerabilities, Jan 2026, including the OpenSSL Foundation statement and the 1998 date · Severity breakdown via SecurityWeek, 28 Jan 2026 · Exploitability, affected software and release date via Datadog Security Labs · Heartbleed and web server share via Netcraft, 7 Apr 2014 · IBM Cost of a Data Breach Report 2026, figures via Secureframe and databreachcost.com · Unit 42 NOVA figures via IBM X-Force · Mandiant, AI risk and resilience in 2026 · LayerX Enterprise AI and SaaS Data Security Report 2025, via Cloud Security Alliance · Cyberhaven, 2026 AI Adoption & Risk Report · Daily and annual paste counts are our arithmetic on Cyberhaven's per-employee average, not a measurement
← Back to Studio Notes © 2026 Studio 111 Labs · Doug Larsen