DEV Community

Cover image for Retire TLS 1.0/1.1 Safely: A Real Protocol & Cipher Audit
Muhammad Ali Mughal
Muhammad Ali Mughal

Posted on

Retire TLS 1.0/1.1 Safely: A Real Protocol & Cipher Audit

Most teams think they turned off TLS 1.0 years ago. Then a scanner, a PCI assessor or a partner's security questionnaire asks for proof, and nobody can say which protocols and cipher suites the server actually accepts today. The fix is not a copied config from a forum. It is a ten-minute audit with openssl s_client, one access-log field that tells you who still needs old protocols, and a config you can explain line by line.

I build PixelValut, a set of free browser-side security tools. I wrote this with help from AI tools and checked every command, output and claim myself. The outputs below come from real handshakes I ran against pixelvalut.com and the public badssl.com test hosts on Nov 16, 2026, using OpenSSL 3.5.

How do you safely retire TLS 1.0 and 1.1?

Direct answer: Test what your server negotiates per protocol with openssl s_client (forcing a permissive client security level), log $ssl_protocol in access logs for a week to find real legacy clients, then allow only TLS 1.2 and 1.3 with ECDHE plus AEAD ciphers. Re-test, and keep the evidence for audits.

Why this still matters in 2026

TLS 1.0 and 1.1 were formally deprecated by RFC 8996 in 2021, and mainstream browsers stopped supporting them back in 2020. So why audit at all?

  • Browsers are not your only clients. ERP connectors, payment terminals, old Java and .NET integrations, embedded devices and monitoring agents often lag years behind. They are exactly the clients that break when you tighten TLS, and exactly the ones nobody remembers.
  • Load balancers and CDNs hide the real config. Your origin might be strict while the edge still accepts TLS 1.0, or the reverse. Each listener needs its own test.
  • Weak TLS 1.2 suites are the quieter problem. A server can refuse TLS 1.0 and still offer CBC-mode or static-RSA key exchange suites under TLS 1.2, which give up forward secrecy.
  • Auditors want evidence, not intent. "We disabled it" is a claim. A dated handshake log is proof.

The trap: your client can lie to you

The first time I ran a protocol test with a modern OpenSSL, this is what came back for TLS 1.0:

$ openssl s_client -connect pixelvalut.com:443 -servername pixelvalut.com -tls1 </dev/null
error:0A0000BF:SSL routines:tls_setup_handshake:no protocols available
New, (NONE), Cipher is (NONE)
Enter fullscreen mode Exit fullscreen mode

That looks like "server refuses TLS 1.0", but read the error: tls_setup_handshake: no protocols available happens before any packet leaves the machine. OpenSSL 3's default security level refuses TLS 1.0 and 1.1 on the client side. I got the identical error against tls-v1-0.badssl.com:1010, a host that exists specifically to accept TLS 1.0. A green result from this test proves nothing about the server.

To test the server, lower the client's security level for the probe:

$ openssl s_client -connect tls-v1-0.badssl.com:1010 -servername tls-v1-0.badssl.com \
    -tls1 -cipher 'DEFAULT@SECLEVEL=0' </dev/null
...
    Protocol  : TLSv1

$ openssl s_client -connect pixelvalut.com:443 -servername pixelvalut.com \
    -tls1 -cipher 'DEFAULT@SECLEVEL=0' </dev/null
error:0A000410:SSL routines:ssl3_read_bytes:ssl/tls alert handshake failure:...:SSL alert number 40
New, (NONE), Cipher is (NONE)
Enter fullscreen mode Exit fullscreen mode

Now the difference is meaningful. SSL alert number 40 is a handshake_failure alert sent by the server: the request went out and the server said no. Same result for -tls1_1. That is the evidence you want.

The full audit, one host at a time

Here is the sequence I run, with the real results for pixelvalut.com:

# What does a normal modern client get?
$ openssl s_client -connect pixelvalut.com:443 -servername pixelvalut.com -brief </dev/null
Protocol version: TLSv1.3
Ciphersuite: TLS_AES_256_GCM_SHA384
Signature type: ecdsa_secp256r1_sha256
Verification: OK
Peer Temp Key: X25519, 253 bits

# Is TLS 1.2 still available for older clients, and with what?
$ openssl s_client -connect pixelvalut.com:443 -servername pixelvalut.com -tls1_2 -brief </dev/null
Protocol version: TLSv1.2
Ciphersuite: ECDHE-ECDSA-AES256-GCM-SHA384
Peer Temp Key: X25519, 253 bits

# Does the server accept CBC-mode suites under TLS 1.2?
$ openssl s_client ... -tls1_2 -cipher 'ECDHE-RSA-AES128-SHA:ECDHE-ECDSA-AES128-SHA:AES128-SHA'
...SSL alert number 40

# Does it accept static RSA key exchange (no forward secrecy)?
$ openssl s_client ... -tls1_2 -cipher 'AES256-GCM-SHA384:AES128-GCM-SHA256'
...SSL alert number 40

# Which TLS 1.3 suites are enabled?
$ openssl s_client ... -tls1_3 -ciphersuites TLS_AES_128_GCM_SHA256
New, TLSv1.3, Cipher is TLS_AES_128_GCM_SHA256
$ openssl s_client ... -tls1_3 -ciphersuites TLS_CHACHA20_POLY1305_SHA256
New, TLSv1.3, Cipher is TLS_CHACHA20_POLY1305_SHA256
Enter fullscreen mode Exit fullscreen mode

Reading it: TLS 1.3 and 1.2 only, ECDHE key exchange with X25519, AEAD ciphers only, CBC and static RSA refused. That is the target state. The public badssl hosts are useful as negative controls: rc4.badssl.com and 3des.badssl.com both ended in a handshake failure from my modern client, which tells you your tool will not silently accept broken ciphers either.

For a quick browser-side view of the certificate, expiry and headers on the same host, I use the PixelValut SSL Checker, and I covered certificate reading in how to check an SSL certificate in 10 minutes. For a full matrix across every suite, testssl.sh or nmap --script ssl-enum-ciphers -p 443 host enumerate everything in one run.

Network cables plugged into a rack switch with green status lights

Image: AI-generated.

Comparison: what each protocol and suite family means

Option Status in 2026 Forward secrecy Who still needs it My call
SSL 3.0 / TLS 1.0 / TLS 1.1 Deprecated (RFC 8996) Depends on suite Very old devices, unpatched integrations Disable
TLS 1.2 + ECDHE + AES-GCM / ChaCha20 Current, widely supported Yes Older OSes, Java/.NET integrations Keep
TLS 1.2 + CBC (-SHA, -SHA256) Legacy, padding-oracle history Yes with ECDHE Rare legacy clients Disable unless logs prove need
TLS 1.2 + static RSA (AES*-GCM-SHA* without ECDHE) Legacy No Very old Java, some appliances Disable
RC4 / 3DES / export / NULL Broken n/a Nobody legitimate Disable everywhere
TLS 1.3 (AES-GCM, ChaCha20-Poly1305) Current (RFC 8446) Always All modern clients Keep, prefer

Find the clients that would break, before they break

Never flip the switch blind. Log the negotiated protocol and cipher for a week first.

Nginx:

log_format tls '$remote_addr [$time_local] "$request" $status '
               '$ssl_protocol $ssl_cipher "$http_user_agent"';
access_log /var/log/nginx/tls.log tls;
Enter fullscreen mode Exit fullscreen mode

Apache (mod_ssl):

CustomLog logs/tls_log "%h %t \"%r\" %>s %{SSL_PROTOCOL}x %{SSL_CIPHER}x \"%{User-Agent}i\""
Enter fullscreen mode Exit fullscreen mode

Then count:

awk '{print $(NF-2)}' /var/log/nginx/tls.log | sort | uniq -c | sort -rn
Enter fullscreen mode Exit fullscreen mode

If anything other than TLSv1.2 and TLSv1.3 shows up, look at the user agent and source IP. In my experience it is usually a forgotten cron job, a monitoring probe or a partner's integration server. Contact the owner with a date before you cut it off. Behind a CDN, use the edge's own logs or settings, because the origin only sees the CDN's connection.

The config I ship

Nginx:

ssl_protocols TLSv1.2 TLSv1.3;
ssl_ciphers ECDHE-ECDSA-AES128-GCM-SHA256:ECDHE-RSA-AES128-GCM-SHA256:ECDHE-ECDSA-AES256-GCM-SHA384:ECDHE-RSA-AES256-GCM-SHA384:ECDHE-ECDSA-CHACHA20-POLY1305:ECDHE-RSA-CHACHA20-POLY1305;
ssl_prefer_server_ciphers off;
ssl_session_timeout 1d;
ssl_session_cache shared:TLS:10m;
Enter fullscreen mode Exit fullscreen mode

Apache:

SSLProtocol -all +TLSv1.2 +TLSv1.3
SSLCipherSuite ECDHE-ECDSA-AES128-GCM-SHA256:ECDHE-RSA-AES128-GCM-SHA256:ECDHE-ECDSA-AES256-GCM-SHA384:ECDHE-RSA-AES256-GCM-SHA384:ECDHE-ECDSA-CHACHA20-POLY1305:ECDHE-RSA-CHACHA20-POLY1305
SSLHonorCipherOrder off
Enter fullscreen mode Exit fullscreen mode

This is close to Mozilla's "Intermediate" server profile; check the current Mozilla SSL Configuration Generator for your server version before you copy anything. Note that ssl_ciphers and SSLCipherSuite only control TLS 1.2 and below. TLS 1.3 suites are configured separately and the defaults are already safe. With only AEAD suites left, letting the client choose (prefer_server_ciphers off) is fine and helps mobile devices pick ChaCha20.

On managed platforms, the same idea is a setting: a CDN "minimum TLS version" option, or a load balancer security policy that names TLS 1.2 as the floor. Set it on every listener, including staging and API subdomains.

Rollout checklist

  1. List every TLS endpoint: apex, www, API, admin, mail (SMTP STARTTLS, IMAPS), staging and load balancer listeners.
  2. Run the per-protocol s_client probes with DEFAULT@SECLEVEL=0 and save the output with a date.
  3. Enable protocol and cipher logging for seven days.
  4. Identify every non-1.2/1.3 client and notify its owner with a cutoff date.
  5. Apply the config on staging, re-run the probes, then roll out to production.
  6. Re-run the probes from outside your network, through the CDN, and against each origin.
  7. Add the probes to a weekly CI or cron job so a server rebuild cannot quietly reintroduce old protocols.
  8. Pair it with HSTS once you are confident; my notes on a safe HSTS preload rollout cover that step.

Failure modes I keep seeing

  • Testing with a modern client and reading "no protocols available" as a pass. Covered above; it is the most common false green.
  • Fixing the origin and forgetting the edge. The CDN or load balancer terminates TLS, so its policy is the one the internet sees.
  • Forgetting mail and non-HTTP ports. SMTP, IMAP and custom TCP services have their own TLS settings. Test them with openssl s_client -starttls smtp -connect mail.example.com:25. For inbound mail, MTA-STS and TLS-RPT add enforcement on top.
  • Disabling TLS 1.2 too. TLS 1.3-only sounds tidy but still cuts off real clients. Keep 1.2 with strong suites.
  • No regression test. A new server image with distro defaults can reopen what you closed.

FAQ

Is TLS 1.2 still secure in 2026?

Yes, with ECDHE key exchange and AEAD ciphers (AES-GCM or ChaCha20-Poly1305). The weak parts of TLS 1.2 are optional suites such as CBC-mode and static RSA. Disable those and TLS 1.2 is a sound fallback next to TLS 1.3.

Why does openssl say "no protocols available"?

Your local OpenSSL refused to offer that protocol before contacting the server, because its default security level blocks TLS 1.0 and 1.1. Add -cipher 'DEFAULT@SECLEVEL=0' for the probe so the result reflects the server's decision.

What does "SSL alert number 40" mean?

It is a handshake_failure alert sent by the server, meaning it found no protocol or cipher it was willing to use with what your client offered. In a downgrade test, that is the result you want to see.

Will disabling TLS 1.0 and 1.1 break my website for visitors?

Current browsers have not supported them since 2020, so normal visitors are unaffected. The risk is non-browser clients such as old integrations, devices and scripts. Log $ssl_protocol for a week to find them first.

Do I need to configure TLS 1.3 cipher suites?

Usually not. The default TLS 1.3 suites in OpenSSL are AES-256-GCM, ChaCha20-Poly1305 and AES-128-GCM, all strong. The ssl_ciphers and SSLCipherSuite directives only affect TLS 1.2 and earlier.

How often should I re-run the audit?

After every server, load balancer or CDN change, and on a weekly schedule. A scripted s_client check takes seconds and catches regressions from new images or provider defaults.

Conclusion

Retiring old TLS is a measurement problem more than a config problem. Probe each protocol with a client that is actually allowed to try, log who still connects with what, give those owners a date, and then ship a short config you can defend. Keep the dated outputs: they turn the next audit question into a two-minute answer.


Follow & Connect

Watch on YouTube: Resume Boost AI · PixelValut · Daily AI News By Ali

Top comments (1)

Collapse
 
parasraju profile image
Paras •

Nice audit, the false-green OpenSSL trap is clever