Almost every organisation now runs TLS 1.2 and TLS 1.3 side by side, and the real question is no longer which one is better but when it is safe to stop accepting TLS 1.2 on each endpoint. TLS 1.3 removes whole classes of weakness by design, and in July 2026 the IETF published RFC 9846 as the current TLS 1.3 specification and RFC 9851, which puts TLS 1.2 into feature freeze. Switching TLS 1.2 off too early still breaks old clients, appliances and inspection tools, and leaving it on forever blocks post-quantum key exchange. This guide compares the two versions as a policy decision: the security gains, the legacy client problem, the impact on TLS inspection and network visibility, and a phased migration with tests and rollback. It is for network, platform and security architects who own TLS policy. The protocol mechanics are in the How TLS Works Guide and the exact server settings in the TLS Server Configuration Baseline Guide; this page does not repeat them.
Key takeaways
- The short answer is prefer 1.3 everywhere, retire 1.2 endpoint by endpoint. BCP 195, as updated by RFC 9852 (July 2026), now makes TLS 1.2 a MAY, requires TLS 1.3 for new protocols that use TLS, and still says TLS 1.3 MUST be preferred, so a well-configured TLS 1.2 is acceptable today, but it is a legacy path with a shrinking future.
- TLS 1.2 is frozen. RFC 9851 (July 2026) says no changes will be approved for TLS 1.2 outside urgent security fixes, and that post-quantum cryptography for TLS 1.2 will not be specified at any time.
- TLS 1.3 security gains are structural, not configurable. Forward secrecy on every full handshake, AEAD-only record protection, encrypted certificates and a downgrade signal mean a misconfigured TLS 1.3 server is still safer than an average TLS 1.2 one.
- Visibility is the real cost. Passive decryption with a copied server RSA key does not work against TLS 1.3, and server certificates are no longer visible on the wire, so monitoring designs built on those must change before TLS 1.2 is turned off.
- Legacy clients set the date, not the server. Windows 10 releases, older Java and embedded devices often cannot negotiate TLS 1.3, so inventory clients from your own handshake logs before you set any cut-off.
Status of both versions in November 2026
The standards picture changed in mid 2026, and several older articles now describe it wrongly. Checked on 14 November 2026:
| Document | Status and date | What it means for policy |
|---|---|---|
| RFC 9846, TLS 1.3 | Proposed Standard, July 2026; obsoletes RFC 8446 and RFC 5246 | The current TLS 1.3 specification. Same version number and backward compatible with RFC 8446; cite RFC 9846 in new policies |
| RFC 8446, TLS 1.3 (2018) | Obsoleted by RFC 9846 | Implementations built to it remain valid TLS 1.3; update document references |
| RFC 5246, TLS 1.2 (2008) | Obsoleted by RFC 9846 as a document | Obsoleting the document is not a ban on the protocol; TLS 1.2 is still permitted by BCP 195 |
| RFC 9851, TLS 1.2 is in Feature Freeze | Proposed Standard, July 2026 | Only urgent security fixes, plus new exporter labels and ALPN IDs; no post-quantum key exchange for TLS 1.2 |
| RFC 9325 (BCP 195) | Best Current Practice, November 2022; updated by RFC 9852 (July 2026) | As updated: TLS 1.2 MAY be supported, TLS 1.3 MUST be supported for new protocols using TLS and MUST be preferred |
| RFC 8996 (BCP 195) | Best Current Practice, March 2021 | TLS 1.0 and 1.1 deprecated; RFC 9846 also forbids negotiating them |
| NIST SP 800-52 Rev. 2 | Final, August 2019; NIST notes (May 2026) it is being reviewed | US government servers and clients must support TLS 1.2 with FIPS-based suites, and TLS 1.3 by 1 January 2024 |
| NCSC, Using TLS to protect data | Current guidance | Use TLS 1.3 where possible; TLS 1.2 still strong against classical attacks with the recommended profiles; plan migration to TLS 1.3 |
RFC 9846 is described in its own Section 1.2 as a minor update. Among other changes it forbids reusing key shares between connections, makes key updates mandatory before key usage limits are reached and caps the number of KeyUpdate messages, adds a general_error alert, and uses more generic key exchange wording so that KEM-based key exchange fits cleanly.
Side-by-side comparison
| Property | TLS 1.2 | TLS 1.3 |
|---|---|---|
| Key exchange | Static RSA, DHE and ECDHE all possible; forward secrecy only if configured | Static RSA and static DH removed; every full handshake uses ephemeral (EC)DHE or a KEM |
| Record protection | CBC with HMAC, RC4 historically, AEAD optional | AEAD only: AES-GCM, ChaCha20-Poly1305, AES-CCM |
| Cipher suite meaning | One string bundles key exchange, authentication, cipher and hash | Suite covers only AEAD and hash; key exchange and signatures negotiated separately |
| Handshake latency | Two round trips for a full handshake | One round trip; optional 0-RTT early data on resumption |
| Handshake privacy | Certificates and most extensions in cleartext | Everything after ServerHello encrypted, including the certificate |
| Version negotiation | Version field in the hello; fallback games were possible | supported_versions extension plus downgrade sentinel in ServerHello.random |
| Compression and renegotiation | Present (should be disabled) | Removed; KeyUpdate replaces rekeying by renegotiation |
| Key derivation | TLS PRF | HKDF-based key schedule |
| Encrypted Client Hello (RFC 9849) | Not available | Available; ECH requires TLS 1.3 or above |
| Post-quantum hybrid key exchange | Will never be specified (RFC 9851) | Standardised in RFC 10024 (X25519MLKEM768 and others) |
| Future changes | Urgent security fixes only | Active development |
Security gains of TLS 1.3, and what a hardened TLS 1.2 still leaves
A TLS 1.2 server configured to the four ECDHE AES-GCM suites recommended in RFC 9325 Section 4.2 is not broken. RFC 9325 says TLS 1.2 is as safe as TLS 1.3 when its mitigations are followed. The difference is that TLS 1.2 is only safe when every one of those mitigations is in place, on every endpoint, and stays in place through upgrades and appliance resets. TLS 1.3 removes the options rather than relying on configuration.
Threat model: what changes per attack class
| Attack class | Hardened TLS 1.2 depends on | TLS 1.3 position |
|---|---|---|
| Retrospective decryption after key theft | Static RSA key exchange disabled on every listener | Not possible to configure static RSA; forward secrecy by design |
| Padding oracles and CBC timing (Lucky Thirteen, POODLE family) | CBC suites removed or encrypt_then_mac negotiated | CBC not available |
| Downgrade to weak suites or groups (FREAK, Logjam) | Export and weak DH removed; TLS_FALLBACK_SCSV where old versions remain | Downgrade sentinel and signed transcript; weak groups not defined |
| Compression attacks (CRIME) | Compression off | Compression removed |
| Renegotiation abuse | Secure renegotiation extension; client renegotiation off | Renegotiation removed |
| Harvest now, decrypt later by a future quantum computer | No fix available, ever | Hybrid ML-KEM groups per RFC 10024 |
| Metadata exposure (certificate, SNI) | Certificate visible; SNI visible | Certificate encrypted; SNI hidden only with ECH |
Background on each named attack is in the Historic TLS Attacks Guide. In ATT&CK terms, the relevant adversary behaviours are T1557 Adversary-in-the-Middle, T1040 Network Sniffing and T1689 Downgrade Attack; TLS 1.3 narrows each of them without eliminating the need for certificate validation.
What TLS 1.3 does not fix
- Certificate validation: a client that skips chain or hostname checks is just as exposed on TLS 1.3. The version does nothing for trust store hygiene.
- 0-RTT replay: early data can be replayed. RFC 9325 Section 3.10 says applications MUST avoid 0-RTT unless an explicit specification says when it is safe; for HTTP see RFC 8470. Leave it off on anything that changes state.
- Weak endpoints: the protocol protects the wire, not a compromised server or a leaked private key; see the Leaked TLS Private Key Response.
Legacy clients: who still cannot speak TLS 1.3
The cut-off date for TLS 1.2 is set by the oldest client you must serve, so client inventory comes before any server change. Typical holdouts:
- Windows Schannel: Microsoft's protocol support table lists TLS 1.3 as enabled on Windows Server 2022, Windows Server 2025 and Windows 11, and not supported on any Windows 10 release. Applications that use the operating system TLS stack on those hosts, including many .NET and PowerShell clients, will only negotiate TLS 1.2. Microsoft warns that enabling TLS 1.3 on earlier Windows versions is not a safe configuration.
- Older language runtimes and libraries: TLS 1.3 arrived in OpenSSL 1.1.1 and in Java 11; applications pinned to older runtimes, or that set a fixed protocol list in code, stay on TLS 1.2.
- Embedded and OT devices: payment terminals, printers, building systems, industrial gateways and medical devices often ship a TLS stack that will never be updated.
- Appliances and middleboxes: load balancers, WAFs, proxies and inspection devices on old firmware may terminate or re-originate only TLS 1.2, which silently caps the version for everything behind them.
- Partner and B2B integrations: payment gateways, EDI and bank file transfers where the counterparty controls the client.
- Machine clients: service accounts, agents and scripts calling APIs. These are non-human identities, and their libraries are often the oldest code in the estate.
The evidence source is your own handshake telemetry: the negotiated version and the client's offered versions per endpoint, joined to client identity (source, user agent, client certificate or API key). Without it, a TLS 1.2 shutdown is guesswork.
Inspection and visibility impact
This is where most TLS 1.3 projects stall. Three monitoring techniques that worked on TLS 1.2 behave differently:
| Technique | TLS 1.2 | TLS 1.3 | What to do |
|---|---|---|---|
| Passive decryption with the server's private key (out-of-band IDS, APM taps) | Works only for static RSA key exchange, which should already be off | Does not work; every session uses ephemeral keys | Decrypt at the termination point (load balancer, proxy) and mirror cleartext, or export session keys from the endpoint under strict control |
| Inline interception proxy (forward proxy, NGFW, SWG) | Works | Works if the proxy fully terminates and re-originates TLS 1.3 sessions | Confirm the device negotiates TLS 1.3 on both legs; otherwise it downgrades every flow to TLS 1.2 |
| Certificate-based classification (reading the server certificate in transit) | Works | Certificate is encrypted | Use SNI, DNS logs and JA4-style fingerprints; plan for SNI loss when ECH is used |
| SNI-based filtering | Works | Works unless ECH is negotiated | Decide an ECH policy for managed endpoints; RFC 9849 notes DNS lookups and IP addresses can still reveal the destination |
Two decisions follow. First, if an inspection device can only speak TLS 1.2, it is effectively a TLS 1.2 policy for every user behind it: upgrade or replace it before you call the estate TLS 1.3 ready. Second, an architecture that relies on passive decryption of server traffic is already incompatible with forward secrecy, so moving it to termination-point visibility is needed whatever you decide about TLS 1.2. The options are compared in the TLS Inspection vs Encrypted Traffic Analysis guide and the TLS Inspection Guide; fingerprinting is in the JA3 and JA4 TLS Fingerprinting Guide.
When to keep TLS 1.2 and when to switch it off
| Endpoint type | Recommended policy | Reason |
|---|---|---|
| Internal service to service traffic you fully control | TLS 1.3 only | You own both ends; no legacy argument |
| New APIs and new applications | TLS 1.3 only from day one | Adding a legacy path later is easier than removing one |
| Public websites for consumers | TLS 1.3 preferred, TLS 1.2 with RFC 9325 suites retained while logs show meaningful TLS 1.2 traffic | Long tail of old browsers and devices |
| Partner and B2B endpoints | TLS 1.2 allowed per named partner with a sunset date in the contract | Counterparty controls the client |
| Embedded, OT and payment devices | TLS 1.2 on a segmented listener with an exception record | Device stacks cannot be upgraded on your timeline |
| Systems needing post-quantum protection for long-lived secrets | TLS 1.3 only with a hybrid group | TLS 1.2 will never get PQC |
| FIPS-constrained systems | TLS 1.3 where the validated module supports it, TLS 1.2 with FIPS suites otherwise | SP 800-52 Rev. 2 requires both to be supported |
Small organisations usually get TLS 1.3 from a CDN, a managed load balancer or a SaaS front end and only need to set the minimum version and check partners. Large organisations need the inventory, exception process and telemetry described below, because thousands of internal listeners and appliances each carry their own defaults.
Worked scenario
A retailer runs a public web shop behind a CDN, an internal API platform, store payment terminals and a forward proxy for staff browsing. Handshake logs show 0.4 per cent of shop traffic on TLS 1.2, mostly old Android devices; all internal API callers support TLS 1.3 except a reporting job on a Windows Server 2019 host; terminals are TLS 1.2 only; the forward proxy re-originates outbound traffic as TLS 1.2. The decision: API platform goes TLS 1.3 only after the reporting job moves to a supported host; shop keeps TLS 1.2 with the four RFC 9325 suites and a review every quarter; terminals stay on a segmented TLS 1.2 listener with an exception tied to the device refresh; the proxy firmware upgrade goes first because it caps every staff connection.
Migration steps
- Inventory listeners and clients. Scan every TLS endpoint externally and internally for supported versions, and turn on negotiated-version logging at load balancers, proxies, API gateways and web servers.
- Enable TLS 1.3 everywhere, alongside TLS 1.2. This is low risk because version negotiation picks the highest common version. Confirm TLS 1.3 is preferred (RFC 9325 3.1.1) and 0-RTT is off.
- Fix the caps. Upgrade inspection devices, load balancers and libraries that force TLS 1.2 on one leg. Re-run scans to confirm both legs negotiate TLS 1.3.
- Move visibility. Replace passive key-based decryption with termination-point logging or inline inspection before any client is forced onto TLS 1.3.
- Harden what remains on TLS 1.2. Restrict it to the RFC 9325 ECDHE AES-GCM suites, extended master secret and secure renegotiation, per the TLS Server Configuration Baseline Guide.
- Pilot TLS 1.3-only by ring. Start with internal service traffic, then new APIs, then low-traffic public hosts. Watch handshake failure rates and help desk tickets for two weeks per ring.
- Handle exceptions formally. Each remaining TLS 1.2 listener gets an owner, a business reason, a segment and a sunset date. NCSC asks for a time-bounded migration plan for anything still on deprecated settings.
- Roll back cleanly. Keep the minimum-version setting in configuration as code so a ring can return to "TLS 1.2 and 1.3" in one change. Failures surface as handshake errors such as protocol_version alerts; see the TLS Handshake Failure Troubleshooting Guide.
Where identity fits
Many TLS 1.2 holdouts are machine clients: service accounts running scheduled jobs, agents, and integrations authenticating with API keys or client certificates. Tie each TLS 1.2 handshake to the identity behind it so the owner can be found and the runtime upgraded. Client certificate authentication works in both versions, but in TLS 1.3 the client certificate is encrypted, which improves privacy and removes another passive visibility source. See the Mutual TLS vs TLS guide.
Detection and measurement
- Negotiated version share: percentage of handshakes on TLS 1.3, per endpoint and overall; target a rising trend and 100 per cent on internal traffic.
- TLS 1.2-only clients: count of distinct clients that never offer TLS 1.3, each mapped to an owner.
- Exception register: number of approved TLS 1.2 exceptions and how many are past their sunset date (target zero overdue).
- Version caps: number of middleboxes or listeners unable to negotiate TLS 1.3 on either leg.
- Post-quantum coverage: share of TLS 1.3 handshakes using a hybrid group such as X25519MLKEM768.
- Handshake failure rate after each ring: protocol_version and handshake_failure alerts compared with the baseline.
Common failure patterns
- Switching off TLS 1.2 from server scans alone: scans show what the server supports, not who connects; the first sign of a missed client is an outage in a partner or batch job.
- Forgetting the inspection leg: a proxy that re-originates TLS 1.2 makes a TLS 1.3 project look complete in server logs while every user is still on TLS 1.2.
- Turning on 0-RTT for speed: early data enabled on an API that accepts state-changing requests opens replay risk.
- Treating "TLS 1.2 obsolete" as "TLS 1.2 forbidden": reading RFC 9846 as a ban leads to rushed shutdowns; BCP 195 still permits TLS 1.2, so plan the retirement.
- Losing monitoring silently: passive decryption tools stop seeing content after the change and nobody notices until an investigation needs it.
- Leaving weak TLS 1.2 suites on during the transition: the legacy path stays open for years, so harden it rather than ignoring it.
Framework mapping
| Framework | Reference | Relevance |
|---|---|---|
| NIST CSF 2.0 | PR.DS-02, PR.PS-01 | Data in transit protection; configuration management of TLS settings |
| NIST SP 800-53 Rev. 5 | SC-8, SC-13, SC-23, CM-7 | Transmission confidentiality, approved cryptography, session authenticity, disabling unneeded protocol versions |
| CIS Controls v8.1 | 3.10 Encrypt Sensitive Data in Transit | Minimum protocol policy |
| ISO/IEC 27001:2022 Annex A | A.8.24 Use of cryptography | Cryptography rules covering protocol versions |
| MITRE ATT&CK v20 | T1557, T1040, T1689 | Interception, sniffing and downgrade behaviours that TLS 1.3 narrows |
Practitioner checklist
- Update policy references from RFC 8446 to RFC 9846 and record RFC 9851's feature freeze.
- Enable TLS 1.3 on every listener and confirm it is preferred over TLS 1.2.
- Log negotiated version and offered versions per endpoint, joined to client identity.
- Find and upgrade every middlebox that caps sessions at TLS 1.2.
- Move visibility from passive decryption to termination-point logging or inline inspection.
- Restrict remaining TLS 1.2 to the RFC 9325 ECDHE AES-GCM suites.
- Make internal and new services TLS 1.3 only, ring by ring, with a one-change rollback.
- Keep 0-RTT off unless the application protocol defines safe use.
- Give each TLS 1.2 exception an owner, segment and sunset date.
- Plan hybrid post-quantum groups, which require TLS 1.3; see the Hybrid Post-Quantum TLS Key Exchange Guide.
Standards and references
- IETF: RFC 9846, The Transport Layer Security (TLS) Protocol Version 1.3
- IETF: RFC 9851, TLS 1.2 is in Feature Freeze
- IETF: RFC 9325, Recommendations for Secure Use of TLS and DTLS (BCP 195)
- IETF: RFC 9852, New Protocols Using TLS Must Require TLS 1.3 (BCP 195)
- IETF: RFC 9849, TLS Encrypted Client Hello
- IETF: RFC 10024, Post-Quantum Traditional Hybrid Key Agreement for TLS 1.3
- NIST: SP 800-52 Rev. 2, Guidelines for the Selection, Configuration, and Use of TLS Implementations
- NCSC: Using TLS to protect data
- Microsoft Learn: Protocols in TLS/SSL (Schannel SSP)
Related NHI Mgmt Group resources: How TLS Works Guide, TLS Server Configuration Baseline Guide, Historic TLS Attacks Guide, Post-Quantum Cryptography Migration Guide.