In ssh-agent in OpenSSH before 10.5, some operations can occur remotely but were intended to occur only locally, including operations that add tokens or use keys. This is caused by misinteraction between agent locking and the session-bind@openssh.com extension.

In ssh in OpenSSH before 10.5, a use-after-free for realloc data can occur if a certain pair of remote-forwarding operations are concurrent.

In sshd in OpenSSH before 10.5, the restrict keyword (in authorizedkeys) was supposed to be applicable to tunnel forwarding but was not.

The AI industry is running a contradiction it hasn’t resolved.

Frontier labs are warning about an imminent software security collapse. Anthropic reported that Claude Mythos found zero-days in codebases hardened by decades of review: a 27-year-old denial-of-service bug in OpenBSD’s TCP SACK implementation, and a 17-year-old remote code execution flaw in FreeBSD’s NFS server that hands an unauthenticated user root (CVE-2026-4747). Mozilla, testing Mythos Preview against Firefox 150, surfaced 271 vulnerabilities, more than ten times what it found in Firefox 148 using Claude Opus 4.6. OpenAI is sounding the exact same alarm, touting GPT-6 Astra saturating ExploitBench while warning that autonomous attack capabilities are scaling faster than human defenders can patch.

To be fair, Anthropic acted on part of this. Project Glasswing puts $100M in credits behind partnerships with Microsoft, Apple, CrowdStrike, and the Linux Foundation.

Glasswing makes sense for foundational open-source packages and mega-cap infrastructure. It does nothing for the other 99% of software developers. The startups, the internal tool builders, and the mid-market engineering teams writing the proprietary software that runs the rest of the world don’t have an organizational sponsor. Yet they are expected to hold off machine-speed exploits without access to the intelligence that found them.

What they get instead are models like Claude Fable or public endpoints of Astra, sitting behind hair-trigger safety classifiers and preemptive capability caps.

Labs are terrified of prompt injection and “I’m an authorized pen tester” jailbreaks, so the filters cast an absurdly wide net. Paste in complex code and ask the model to trace an execution path, analyze memory corruption, or check whether untrusted input can reach a sensitive sink, and the classifier fires immediately.

Security researchers have already documented Claude Code blocking vulnerability work outright, throwing errors about “violative cyber content” that contaminate the entire session and spread to benign follow-up questions. Worse, when the classifier flags an inquiry as security-related, it quietly routes the work to Opus. Opus handles the analysis, and you get degraded work compared to what Fable would have delivered.

Anthropic built Mythos as a model tier above Opus and published what that difference looks like in practice: ten times the findings on comparable Firefox releases.

So when Opus finishes auditing your code and reports nothing exploitable, what have you actually learned? Only that there are no bugs an Opus-class model can find. Anthropic’s own numbers prove that is a fraction of what is actually there. You haven’t verified your code is secure; you’ve verified it survived a search one tier below the frontier, and you were never told the search happened at that tier.

That ceiling won’t hold. Every audit that passes at the Opus level is a bet that nothing Mythos or Astra-class will ever be pointed at your attack surface. But frontier engines already exist, and ablated open-weight models in agentic loops are rapidly converging on the exact same workflows, without a classifier deciding which tier is appropriate for them. On a long enough timeline, that bet is guaranteed to lose.

The gatekeeping rests on a misunderstanding of how software security actually works.

Defense is downstream of offense. You cannot verify a fix you cannot attack. A scanner that lists 300 theoretical warnings without the ability to chain an exploit path is useless. It just breeds alert fatigue. Real remediation means proving the vector is reachable, building the proof of concept, shipping the patch, and running the exploit again to confirm the vector is dead.

Neuter the model’s offensive reasoning, and you neuter the audit.

Meanwhile, nobody attacking your systems is applying to Project Glasswing or arguing with commercial API guardrails.

They run local open-weight models: ablated, fine-tuned on exploit repositories, with the guardrails stripped out. Open weights still trail closed frontier models on raw benchmarks, but benchmark scores don’t decide an engagement. An attacker doesn’t need one omniscient model. They just need uncensored weights in an agentic loop with a fuzzing harness behind it, and no safety classifier killing the context window on attempt four hundred.

That leaves three distinct tiers:

Incumbents, who get vetted private access to the frontier tier through closed programs like Glasswing. The 99% of developers, who get models like Fable or Astra that trip alarms, refuse prompts, or quietly hand security work down to lower tiers without telling anyone. Attackers, who get unconstrained offensive intelligence pointed straight at the attack surface, running at machine speed.

Telling developers a cyber tsunami is coming while confiscating their lifeboats is not responsible stewardship. It is security theater. If the labs believe machine-speed cyber warfare is already here, throttling the people trying to inspect their own code is a strange way to act on that belief. Give everyday builders the same offensive firepower and let them tear their own systems apart before someone else does.

Originally posted at: https://eddiemissri.substack.com/p/the-asymmetric-disarmament-of-ai

A vulnerability was found in OpenSSH when the VerifyHostKeyDNS option is enabled. A machine-in-the-middle attack can be performed by a malicious machine impersonating a legit server. This issue occurs due to how OpenSSH mishandles error codes in specific conditions when verifying the host key. For an attack to be considered successful, the attacker needs to manage to exhaust the client's memory resource first, turning the attack complexity high.

A flaw was found in OpenSSH. A malicious SSH server can exploit a double free vulnerability in the Diffie-Hellman Group Exchange (DH-GEX) client path. This occurs during FIPS (Federal Information Processing Standards) mode known-group validation when the client processes attacker-controlled DH-GEX group parameters. Successful exploitation leads to client-side process termination, resulting in a Denial of Service (DoS).

In OpenSSH before 10.3, a file downloaded by scp may be installed setuid or setgid, an outcome contrary to some users' expectations, if the download is performed as root with -O (legacy scp protocol) and without -p (preserve mode).

An integer underflow was found in the popt library when formatting help text for option tables that exceed the terminal width. A local user who can cause an application to print help under those conditions may cause that application to crash or fail to display help, resulting in a denial of service of the affected application.

A security regression (CVE-2006-5051) was discovered in OpenSSH's server (sshd). There is a race condition which can lead sshd to handle some signals in an unsafe manner. An unauthenticated, remote attacker may be able to trigger it by failing to authenticate within a set time period.

Last updated 30 June 2026

A flaw was found in OpenSSH. A local unprivileged attacker on a Linux client host can hijack client-side X11 forwarding connections. This is possible by pre-binding the preferred abstract X socket name when X11 forwarding is enabled and a local UNIX-domain X socket is used. A successful attack can compromise the confidentiality of forwarded X11 traffic, including sensitive window contents and input, and may allow some manipulation of the forwarded session.

A flaw was found in OpenSSH. This vulnerability, a heap out-of-bounds read, occurs during the cleanup of GSSAPI (Generic Security Service Application Programming Interface) indicators when a trailing NULL termination is missing in the auth-indicators array. A remote attacker, under specific configurations involving GSSAPI authentication and a Kerberos environment, could exploit this to cause the SSH authentication path to crash or abort. This leads to a denial of service (DoS), impacting the availability of the SSH service.

Last updated 13 July 2026

Last updated 13 July 2026

In OpenSSH before 10.3, command execution can occur via shell metacharacters in a username within a command line. This requires a scenario where the username on the command line is untrusted, and also requires a non-default configurations of % in sshconfig.

OpenSSH before 10.3 can use unintended ECDSA algorithms. Listing of any ECDSA algorithm in PubkeyAcceptedAlgorithms or HostbasedAcceptedAlgorithms is misinterpreted to mean all ECDSA algorithms.

OpenSSH before 10.3 omits connection multiplexing confirmation for proxy-mode multiplexing sessions.

Last updated 10 July 2026

Last updated 13 July 2026

Last updated 13 July 2026

internal-sftp in sshd in OpenSSH before 10.4 recognizes only the first 9 command-line arguments, which can be important if a later command-line argument would have helped to ensure the intended security properties of an SFTP connection.

In sshd in OpenSSH before 10.4, DisableForwarding=yes was supposed to take precedence over PermitTunnel=yes, but did not.

Last updated 13 July 2026

Summary

Terrapin is a prefix truncation attack targeting the SSH protocol. More precisely, Terrapin breaks the integrity of SSH's secure channel. By carefully adjusting the sequence numbers during the handshake, an attacker can remove an arbitrary amount of messages sent by the client or server at the beginning of the secure channel without the client or server noticing it.

Mitigations

To mitigate this protocol vulnerability, OpenSSH suggested a so-called "strict kex" which alters the SSH handshake to ensure a Man-in-the-Middle attacker cannot introduce unauthenticated messages as well as convey sequence number manipulation across handshakes.

Warning: To take effect, both the client and server must support this countermeasure.

As a stop-gap measure, peers may also (temporarily) disable the affected algorithms and use unaffected alternatives like AES-GCM instead until patches are available.

Details

The SSH specifications of ChaCha20-Poly1305 (chacha20-poly1305@openssh.com) and Encrypt-then-MAC (-etm@openssh.com MACs) are vulnerable against an arbitrary prefix truncation attack (a.k.a. Terrapin attack). This allows for an extension negotiation downgrade by stripping the SSHMSGEXTINFO sent after the first message after SSHMSGNEWKEYS, downgrading security, and disabling attack countermeasures in some versions of OpenSSH. When targeting Encrypt-then-MAC, this attack requires the use of a CBC cipher to be practically exploitable due to the internal workings of the cipher mode. Additionally, this novel attack technique can be used to exploit previously unexploitable implementation flaws in a Man-in-the-Middle scenario.

The attack works by an attacker injecting an arbitrary number of SSHMSGIGNORE messages during the initial key exchange and consequently removing the same number of messages just after the initial key exchange has concluded. This is possible due to missing authentication of the excess SSHMSGIGNORE messages and the fact that the implicit sequence numbers used within the SSH protocol are only checked after the initial key exchange.

In the case of ChaCha20-Poly1305, the attack is guaranteed to work on every connection as this cipher does not maintain an internal state other than the message's sequence number. In the case of Encrypt-Then-MAC, practical exploitation requires the use of a CBC cipher; while theoretical integrity is broken for all ciphers when using this mode, message processing will fail at the application layer for CTR and stream ciphers.

For more details see https://terrapin-attack.com.

Impact

This attack targets the specification of ChaCha20-Poly1305 (chacha20-poly1305@openssh.com) and Encrypt-then-MAC (-etm@openssh.com), which are widely adopted by well-known SSH implementations and can be considered de-facto standard. These algorithms can be practically exploited; however, in the case of Encrypt-Then-MAC, we additionally require the use of a CBC cipher. As a consequence, this attack works against all well-behaving SSH implementations supporting either of those algorithms and can be used to downgrade (but not fully strip) connection security in case SSH extension negotiation (RFC8308) is supported. The attack may also enable attackers to exploit certain implementation flaws in a man-in-the-middle (MitM) scenario.

Accessibility. A privacy issue was addressed with improved private data redaction for log entries.

OpenBSD before commit 6a23123 (2026-06-18) contains an out-of-bounds read vulnerability in the mplsdoerror function within sys/netmpls/mplsinput.c that allows remote attackers to disclose kernel stack memory by sending crafted MPLS frames with 16 labels and no Bottom-of-Stack bit set.

Accessibility. A privacy issue was addressed with improved private data redaction for log entries.

Last updated 24 July 2024

Last updated 11 February 2025

sppppapinput in sys/net/ifspppsubr.c in OpenBSD before 076e2b1 allows authentication bypass via certain zero values for lengths.