Industries / Technology / Software and SaaS
Cybersecurity
Cybersecurity is the work of protecting computers, networks and the data on them from people who try to break in, steal, spy or cause damage.
1970
A Defense Science Board task force chaired by Willis Ware of the RAND Corporation completed "Security Controls for Computer Systems", a classified report that said of computer security "This is a new field" and called itself "the first major attempt to codify its principles".
49
Milestones achieved across 7 domains.
11
To watch for: announced, not yet confirmed outside.
14
Open deliverables. Nobody has shipped them.
In one minute
Where it started
Computer security became a field of its own once many people shared one computer and each needed their files kept from the others. In 1970 a US defense task force led by Willis Ware set out the controls such shared machines needed, and in 1972 James Anderson warned the US Air Force that security could not be added to a system after it was built. The first tools were passwords, audit logs and careful reviews of how a system was designed.
Where it stands
Defenders now work from shared public lists: one ID for each known flaw, a US government catalog of flaws attackers are known to use, and MITRE's map of what attackers do once inside. Sign-in that resists phishing and encryption meant to survive future quantum computers are spreading through the largest online services, while controlled trials find that phishing training changes little. AI models now find flaws, write phishing emails and sort security alerts, and they do it for attackers and defenders alike.
What is open
Much of the open work is measurement by someone outside: whether training, AI-written fixes, detection rules and memory-safe languages do what their makers say. The rest is inventories nobody keeps, such as which software makers published memory safety plans, which network devices keep the records investigators need, and which plug-ins for AI agents ask for more access than they use. All of it can be done with public data, open-source code and test machines the builder owns.
Domains
AchievedWatch forOpen
Domain
Progress
Last achieved
Watch for
Open
Finding and fixing weaknesses
Finding the flaws in software that is already in use, deciding which to fix first, and getting the fixes installed.
7 · 2 · 2
Last achievedNIST stops enriching every flaw2026
Watch forAI tops a bounty leaderboard
2
openIdentity and access
Deciding who and what may use a computer system, and checking that each is who it claims to be.
7 · 1 · 2
Last achievedStolen passwords, no second factor2024
Watch forPasskeys: one sign-in in three
2
openEncrypting data
Scrambling data so that only the intended people can read it, and proving it has not been altered, whether it is stored or crossing a network.
7 · 2 · 2
Last achievedStandards for the quantum era2024
Watch forTwo thirds of traffic post-quantum
2
openDetecting intrusions
Noticing an attack while it is under way, from the records computers keep, the traffic on a network and the behavior of machines and accounts.
7 · 1 · 2
Last achievedSigned update hides a spy2020
Watch forAI triage that matches analysts
2
openResponding to incidents
Containing an attack once it is found, getting systems running again, and telling the people and authorities who need to know.
7 · 2 · 2
Last achievedMost victims stop paying2025
Watch forA ban on paying ransoms
2
openSecuring software before release
Designing, writing and building software so that fewer flaws reach the people who use it.
7 · 2 · 2
Last achievedA worm in npm2025
Watch forA thousandfold fewer memory flaws
2
openTraining people
Teaching staff to spot and report tricks such as phishing, emails that pose as a trusted sender to get someone to click a link or give up a password, and giving defenders practice at running security under attack.
7 · 1 · 2
Last achievedAI writes lures like experts2026
Watch forBig drops, no control group
2
open
Advances to build with
2 research notes

Gemini 4 Carbon
Cyber defenders first
Google says cyber defenders got Gemini 4 Carbon first, and its 88.2% flaw-finding score is Google's own run on a 183-task subset with guardrails off.

GitHub Copilot Fusion
Effort, not model
GitHub Copilot Fusion picks the AI models for each request, and in a test GitHub paid for, its top setting scored 2 points below a rival at about half the cost.
Open in cybersecurity
3 of 14 open deliverables
Open · Finding and fixing weaknesses
Product data for the backlog
An open pipeline that drafts the missing affected-product data, CPE names with version ranges, for CVE records that NIST has marked "Not Scheduled". An AI model reads the CVE text and the vendor's advisory and drafts the entry; a person checks each draft before it is published as open data that scanners can load.
Built with what exists
the CVE List in JSON (the CVEProject/cvelistV5 repository on GitHub); NIST's CPE dictionary and NVD API; CISA's Vulnrichment repository; a large language model such as Claude or Gemini; Python; OWASP Dependency-Check, an open scanner that matches software to CPE names, to test that the output works.
Open · Identity and access
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.
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").
Open · Encrypting data
Post-quantum defaults, distro by distro
A table of whether the current releases of about 10 major Linux distributions, among them Debian, Ubuntu, Fedora, Red Hat Enterprise Linux, SUSE, Alpine and Arch, agree a post-quantum hybrid key exchange by default in their packaged OpenSSH, OpenSSL, GnuTLS, NSS, curl, nginx, Apache, Go and Java, and in VPN software such as strongSwan. Each cell rests on a test in which both ends of the connection run on the tester's own machines and the key-exchange method actually agreed is recorded. Rerun the tests whenever packages change.
Built with what exists
Docker or Podman images of each distribution; openssl s_client and s_server, gnutls-cli, ssh -vv, and short Go and Java test programs; tshark to record each handshake; GitHub Actions to rerun the tests.
Where it sits
Broader
In outline