BRENNEN

Industries / Technology / Software and SaaS / Cybersecurity

Identity and access

Deciding who and what may use a computer system, and checking that each is who it claims to be.

7

Milestones achieved, from 1962 to 2024.

1

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 1962

    Passwords, then password theft

    MIT's Compatible Time-Sharing System (CTSS) let many people use one computer from separate terminals and gave each user a password to lock their own files. Most of the computer historians Wired asked in 2012 said the first computer password probably came from CTSS. In spring 1962 Allan Scherr, a PhD researcher who wanted more than his four hours a week, had the system print its password file and took the printout, which Wired calls the earliest documented case of password theft. Later a software bug swapped two files and printed the whole password file on every terminal at login, an incident Robert Morris, who was using the system, recorded in 1979.

    LessonA file that holds every password in readable form will leak, by theft or by accident.

    wired.comrist.tech.cornell.edu

    Credit

    Fernando Corbató and the CTSS team at MIT

  2. 1979

    Passwords stored scrambled

    Robert Morris and Ken Thompson described how Unix keeps only a scrambled form of each password, made by a one-way function, so the password file can be read without revealing the passwords. They made the function slow on purpose, running the DES cipher 25 times, and added a 12-bit random number, the salt, to each password, which multiplies the work of testing guesses against a whole file by 4,096. In their test of 3,289 real passwords, 2,831, or 86%, were short strings or words from dictionaries and name lists, and a five-minute dictionary run found about a third.

    LessonPeople choose guessable passwords, so stored passwords have to be slow to test.

    rist.tech.cornell.edu

    Credit

    Robert Morris and Ken Thompson, Bell Laboratories

  3. 2014

    Trust moves off the network

    Companies guarded internal systems with a firewall around the company network and trusted whoever was inside. Google's BeyondCorp paper said an attacker who gets through that perimeter can easily reach the privileged internal network, and that Google was removing the privileged intranet and moving its corporate applications onto the internet. In August 2020 NIST, the US standards agency, published the approach as "zero trust": "no implicit trust" granted to users or devices "based solely on their physical or network location".

    research.googlecsrc.nist.gov

    Credit

    Rory Ward and Betsy Beyer, Google; Scott Rose, Oliver Borchert, Stu Mitchell and Sean Connelly, authors of NIST SP 800-207

  4. 2017

    Stop forcing password changes

    NIST's digital identity guideline SP 800-63B, published in June 2017, told services they "SHOULD NOT" make people change passwords on a schedule or follow rules about mixing character types, and "SHALL" check new passwords against lists of common or breached ones. Its successor, SP 800-63-4, which replaced it on 1 August 2025, turned both "SHOULD NOT"s into "SHALL NOT" and requires a phishing-resistant authenticator at its highest assurance level.

    pages.nist.govpages.nist.gov

    Credit

    NIST

  5. 2018

    Security keys stop phishing

    From early 2017 Google required its more than 85,000 employees to sign in with a physical security key, a small device built to the FIDO U2F standard that the user plugs in and presses, in place of passwords and one-time codes. In July 2018 a Google spokesperson told the journalist Brian Krebs that there had been "no reported or confirmed account takeovers" since. In May 2022 Apple, Google and Microsoft committed to support passkeys, FIDO sign-in credentials that a person's devices can share, so ordinary accounts could get the same protection without a separate key.

    krebsonsecurity.comfidoalliance.org

    Credit

    Google; the FIDO Alliance and W3C

  6. 2023

    A stolen key opens mailboxes

    In May and June 2023 a China-linked group that Microsoft calls Storm-0558 read the Microsoft Exchange Online email of 22 organizations and 503 related personal accounts, including that of the US Commerce Secretary. It forged sign-in tokens with a Microsoft signing key made in 2016 for consumer accounts, which was due to be retired in March 2021 but was not, and a flaw in how tokens were checked let those consumer tokens open business mailboxes. The US State Department noticed first, through a custom alert on a mailbox-access log that came only with a premium license. In March 2024 the US Cyber Safety Review Board called the intrusion preventable, the result of "a cascade of security failures at Microsoft", and found that Microsoft still did not know how the key had been stolen.

    LessonOne signing key can stand in for every password and second factor it vouches for, so keys need an inventory, a retirement date that is kept, and checks on where they are accepted.

    cisa.gov

    Credit

    the Cyber Safety Review Board, which reviewed it

  7. 2024

    Stolen passwords, no second factor

    A group that Mandiant tracks as UNC5537 took data from customer accounts at Snowflake, a cloud data company, by signing in with usernames and passwords that infostealer malware had copied from people's computers, the earliest in November 2020. Mandiant and Snowflake notified about 165 possibly exposed organizations. None of the accounts used had multi-factor authentication (a second proof besides the password); at least 79.7% had credentials exposed before; some passwords had not been changed for as long as four years; and Mandiant found no evidence that Snowflake's own systems had been breached.

    LessonA stolen password keeps working for years unless it is changed or backed by a second factor.

    cloud.google.com

    Credit

    Mandiant, part of Google Cloud, which investigated

Watch for

  1. Passkeys: one sign-in in three

    The FIDO Alliance's Passkey Index, built from confidential data from 15 large consumer services including Amazon, Google, Microsoft, PayPal and WhatsApp, reports that 36% of their successful sign-ins used a passkey in 2026, against 26% across a smaller group of nine in 2025. It says passkey sign-ins succeed 93% of the time against 78% for passwords and traditional multi-factor methods, and that all eight participants that measure phishing saw fewer successful attacks after adding passkeys. The figures are unweighted averages, and no figure is attributed to any one company.

    Achieved whena party outside the alliance, or the participants themselves under their own names, publish sign-in data that shows a similar passkey share and a fall in successful phishing.

    fidoalliance.org

    Claimed by

    FIDO Alliance, October 2026

Open

  1. 01

    Permission audit of MCP servers

    A public table of the 100 most-installed open-source MCP servers, the plug-ins that let AI agents use outside services such as GitHub, Slack or Google Drive through the Model Context Protocol. For each, list the credential it asks for, the permissions (scopes) that credential carries, and the narrowest permissions its tools actually need, worked out from its source code and confirmed by running its own tests against a throwaway test account with only those permissions. Send each project a change where the two differ.

    The brief
    Credit · unclaimed
  2. 02

    Breached-password checks in open source

    A table of the 50 most-used open-source applications and web frameworks that come with their own password sign-in, such as WordPress, Drupal, Django, Devise for Ruby on Rails, Laravel, Keycloak, Nextcloud, GitLab, Mastodon and Discourse. For each default install, record whether it checks new passwords against lists of breached passwords, and whether it still imposes scheduled changes or character-mix rules that NIST now says services "SHALL NOT" require. Then add a breached-password check to five that lack one, switched off by default if their maintainers prefer.

    The brief
    Credit · unclaimed

Open 01 · the brief

Permission audit of MCP servers

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A public table of the 100 most-installed open-source MCP servers, the plug-ins that let AI agents use outside services such as GitHub, Slack or Google Drive through the Model Context Protocol. For each, list the credential it asks for, the permissions (scopes) that credential carries, and the narrowest permissions its tools actually need, worked out from its source code and confirmed by running its own tests against a throwaway test account with only those permissions. Send each project a change where the two differ.

Why it is still open

Astrix Security's 2025 study of more than 5,200 open-source MCP servers counted credential types: 88% need credentials, 53% use long-lived static API keys or personal access tokens, and 8.5% use OAuth. It did not compare the permissions each server asks for with those it needs. A 2026 study in China's Journal of Computer Research and Development analyzed 200 servers in detail and found that each calls on average 40.74 privileged functions of outside services, and that 52% of the permission checks rest on API tokens and passwords; its abstract reports averages, not the narrowest permission each server could use. OWASP lists "Privilege Escalation via Scope Creep" second in its MCP Top 10, still in beta. NIST's National Cybersecurity Center of Excellence asked in February 2026 for comments on a planned project to show how identity standards apply to AI agents; it is a concept paper, not a finished guide. A November 2026 search of the web and GitHub found no per-server comparison.

nhimg.orgcrad.ict.ac.cnowasp.orgnist.gov

Built with what exists

the official MCP Registry and GitHub to pick servers; each provider's published permission lists, such as GitHub's fine-grained token permissions and Google's OAuth scopes; free-tier test accounts; Docker; an AI coding agent such as Claude Code or GitHub Copilot Fusion to read each server's code and draft rows (see the Note "Effort, not model"); a sandbox that keeps keys outside the agent's reach, such as NVIDIA OpenShell (see the Note "Stopped in milliseconds").

  1. First step

    Take the 10 most-starred servers that use a GitHub token and list, tool by tool, which GitHub API calls each makes and the narrowest fine-grained permission that allows them.

  2. Done looks like

    100 servers in the table, each row linking to the code lines that justify its minimum permissions; the server's own tests pass with the narrowed credential for at least 80 of them; a second person rechecks 10 rows and agrees on at least 9; a count of upstream changes accepted.

Open 02 · the brief

Breached-password checks in open source

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A table of the 50 most-used open-source applications and web frameworks that come with their own password sign-in, such as WordPress, Drupal, Django, Devise for Ruby on Rails, Laravel, Keycloak, Nextcloud, GitLab, Mastodon and Discourse. For each default install, record whether it checks new passwords against lists of breached passwords, and whether it still imposes scheduled changes or character-mix rules that NIST now says services "SHALL NOT" require. Then add a breached-password check to five that lack one, switched off by default if their maintainers prefer.

Why it is still open

NIST has told services to check new passwords against breached and common passwords since 2017. Projects review themselves one at a time: Drupal's own review, updated in September 2025, says its core does not check passwords against lists of poor choices and leaves that to add-on modules. A November 2026 search of the web and GitHub found add-ons and single-project reviews but no table across projects.

drupal.orgpages.nist.gov

Built with what exists

local Docker installs of each application; Have I Been Pwned's Pwned Passwords data, which can be downloaded for offline checks; Python for the table.

  1. First step

    Install WordPress, Django and Keycloak on your own machine, try to give a new account the password "password123" in each, and record which accept it.

  2. Done looks like

    50 applications, each finding linked to a line of code, a setting or a documentation page and to a test on a local install; at least five accepted pull requests; a second person repeats 10 of the local tests and gets the same results.