Is FFmpeg safe? A run of subtitle, audio & video flaws (CVE-2026-64830 and more) — even 8.1.2 needs another update

"Is FFmpeg safe to use?" More and more people are searching that about the go-to software used worldwide to convert video and audio. The short answer: FFmpeg itself is trustworthy free software, but running an old version is dangerous. Serious vulnerabilities (software flaws) have piled up through 2026, and a single crafted video or subtitle file can be enough to take over a PC or server.

On July 23, 2026, four new flaws rated 8.8 (high) were disclosed together as part of that wave: subtitle parsing (CVE-2026-64830), hardware video playback (CVE-2026-64831 / CVE-2026-64832), and audio parsing (CVE-2026-64835). The targeted spots differ, but each, if exploited, can let an attacker run code. The catch: none is fixed by June's release 8.1.2. Even if you updated to 8.1.2 in June, you need to check for updates again.

This article answers "is FFmpeg actually safe," explains the four new July flaws, how they relate to CVE-2026-8461 flagged in June, how to tell whether you are affected, and how to keep using FFmpeg safely.

The four new flaws disclosed in July 2026

First, the newest concern: the four July flaws. All are memory-handling bugs (buffer overflows and a double free)—processing a crafted file makes FFmpeg write or free where it should not, and in the worst case run the attacker's code. Severity is on a 10-point scale: 9+ is "critical," 7–8 is "high."

CVE-2026-64830: crash memory by feeding it a subtitle file

The first is a flaw in parsing the subtitle format "VobSub" (used on DVDs; a subtitle is a pair of .sub / .idx files). According to NVD (the U.S. vulnerability database), packing more "stream IDs" (numbers that index the subtitles) than expected into the file makes FFmpeg write past the box it allocated (a heap out-of-bounds write, CWE-787). The result can be a crash or, in a worse case, execution of the attacker's code.

The affected range is very wide—2.1 through 8.1.2, covering many long-running setups. Subtitles are often loaded automatically alongside a video, so "I played/converted a video and the attached subtitle got processed too" can satisfy the condition without the user realizing it. The fix is in FFmpeg's development tree (fix pull request #23657) and is applied via OS update packages as described below.

CVE-2026-64831: a stack overflow in Vulkan HEVC playback

The second is a flaw in playing/processing the newer video codec "HEVC (H.265)" on hardware via the drawing technology Vulkan. Per NVD, planting an abnormal value exceeding the limit into an internal setting (a parameter for HEVC's number of layers) overflows the "stack," a region where the program temporarily holds values (a stack buffer overflow, CWE-121). Because this can be embedded in various video containers FFmpeg supports, such as MP4 or MKV, it can masquerade as an ordinary video file.

The metrics read "network, no authentication, user interaction required (AV:N/AC:L/PR:N/UI:R)"—in other words, it triggers as soon as someone plays or processes the attacker's video. The affected range is 8.0 through 8.1.2. Environments not using this hardware-decoding path are less exposed, but reliably ruling yourself out is hard, so updating is the safe choice. The fix commit is in the development tree.

CVE-2026-64832: a double free in NVIDIA hardware playback

The third is a flaw in "NVDEC," the mechanism that plays/converts video on hardware using NVIDIA GPUs. According to NVD, in the error path when decode working surfaces run out, the same management data is freed twice in a row (a double free, CWE-415). Freeing already-freed memory corrupts the program's internal state and can lead, via a crafted video, to memory corruption or code execution. The affected range is wide—4.4 through 8.1.2—and the fix commit is in the development tree. It mainly concerns conversion servers and apps that use NVIDIA GPU hardware decoding (NVDEC).

CVE-2026-64835: out-of-bounds access via ADX audio files used in games

The fourth is in audio parsing, not video. It affects the decoder for "ADX" (extensions like .adx / .aax, a format used mainly for game audio). Per NVD, when the channel layout (the arrangement of sounds such as left/right) changes mid-stream, the decoder fails to update its internal state correctly and reads/writes outside the box it prepared (out-of-bounds reads and writes, CWE-787). Processing a crafted ADX/AAX file can lead to memory corruption or code execution. Its range is also 4.4 through 8.1.2, and the fix is in the development tree (fix pull request #23659).

The key point is that all four flaws affect versions up to and including 8.1.2, so moving to June's fixed release 8.1.2 alone does not close them. No numbered stable release including these fixes had been published at disclosure, so apply your OS/distribution's security update package once it ships, or—if you build FFmpeg yourself—update to a development build that contains the fix commits (details below).

Is FFmpeg safe to use?

FFmpeg is free, open-source software (anyone can inspect it, and it's free to use) running behind video services, streaming, and editing software worldwide. It is a long-used, trustworthy tool—not something to avoid as "dangerous." For everyday use converting videos you prepared yourself, there is no need to be overly afraid.

What separates safe from unsafe is how you use it. Risk rises in two cases. One is running an old version without updating. The other is processing video or subtitle files of unknown origin. Many of FFmpeg's weak spots hide in the "decoders" and "parsers" that read various formats. Attackers send files crafted to trip up that reading. Conversely, if you apply the latest fixes and only handle trusted files, you avoid most of the danger.

The problem is that FFmpeg vulnerability reports have noticeably increased in 2026. One reason is the spread of AI-assisted vulnerability discovery: in June 2026 an AI agent was reported to have automatically analyzed FFmpeg's roughly 1.5 million lines of code and found 21 previously unknown vulnerabilities (zero-days). So it is not that "FFmpeg suddenly became dangerous," but that previously undiscovered weak spots are now surfacing one after another. That makes frequent updating more of a safety key than ever. All four July flaws and June's CVE-2026-8461 are concrete examples in that wave.

What is FFmpeg, and why does it concern so many people?

FFmpeg is free software for converting video/audio files to other formats and for playing and recording them. Rather than a standalone app, it is the "unsung workhorse" embedded as a component inside many apps and services.

For example, the auto-conversion of uploaded videos on sharing sites, thumbnail generation for streaming services, desktop video players, and recording/streaming software all run FFmpeg. Many tools are built on it, such as the popular Mac video player IINA and the home media server Jellyfin. Even if you never think "I'm using FFmpeg," you almost certainly run it indirectly. That is why "is FFmpeg safe" concerns so many people.

All of these flaws sit in the part of FFmpeg that reads or plays files of various formats. The four July flaws are in "subtitle parsing," "Vulkan HEVC playback," "NVIDIA hardware playback," and "ADX audio parsing," and June's is in "reading the MagicYUV recording format"—different spots, but they share the structure of "the entry point that interprets an incoming file becomes the weakness."

Who targets this, and what do they walk away with?

"Getting taken over just by opening a file" may sound remote from your life. But what makes this run of flaws frightening is that the entry point is "a single ordinary video or subtitle file"—something everyone receives daily. Here is who steps you onto this landmine, and why.

The attackers are, for instance, people who deliberately post broken videos or subtitles to sharing sites and social media, or send a "here's a sample clip" file to a business video-conversion system. What they want is what that conversion server holds: members' registered emails and passwords, credit-card numbers used for payment, unreleased footage, internal authentication keys. The moment one crafted file flows into the conversion process, a path opens on their side to pull out the contents of that box.

Technically, an attacker does reconnaissance to guess "this service auto-processes uploaded files with FFmpeg," then prepares a broken file that reacts only to that processing. Because user uploads are normally trusted and processed as-is, the attacker can use the legitimate upload form itself as the entry point. Processing runs before the server side notices, and with bad luck the server's control is seized. Once the entry point is held, other users' data running on the same mechanism is exposed in a cascade.

The "8.8 severity" figure is only a gauge of technical seriousness. For the operator, what is truly lost is the users' personal data entrusted to them and the never-recoverable trust that "that company's service is safe." Reaching that far from processing a single file is exactly the danger of this class of flaw—so applying the latest fix before exploitation spreads is what matters most.

Are you affected, and how do you update?

FFmpeg is used so broadly that risk varies by your role. Check your situation in the table below. Because the four July flaws are unfixed even in 8.1.2, the "what to do" is to update to the latest fixed build.

Because FFmpeg is often embedded as a component in OSes and apps, it is not easy to know "where FFmpeg is installed on my machine." Below are the sources and update methods by OS/distribution (the packaging form, such as Ubuntu). Since no new stable release existed at disclosure for the July flaws, the reliable move is to check whether each vendor has shipped a security update.

Note that CVE-2026-8461 (see below), flagged in June, is closed by updating to the fixed release 8.1.2. The four July flaws add a second step requiring a fix beyond that—think of it as a two-stage update.

What was the June flaw, CVE-2026-8461 (MagicYUV)?

Here we revisit CVE-2026-8461, published June 18, 2026, which we flagged earlier. Severity is 8.8 (high), and it affects all versions before 8.1.2; updating to the fixed release 8.1.2 closes it. It is worth checking alongside the four July flaws.

The problem was in FFmpeg's parsing of the "MagicYUV" video format—a lossless format used for game-capture recording and video production. The flaw type is an "out-of-bounds write" (CWE-787), where the program writes outside the memory "box" it prepared. Specifically, it was in the MagicYUV parsing file (libavcodec/magicyuv.c) inside the video component "libavcodec."

Feeding it a malformed MagicYUV video with crafted internal sizes causes writes to memory that should be off-limits, leading to a crash (DoS) at best and remote code execution at worst. The official CVE record notes it "causes denial of service and may in some cases be exploited for remote code execution." It was reported by the security research team at JFrog, and CISA's initial assessment found no sign of actual exploitation. The fix is public in fix pull request #23159.

How to keep using FFmpeg safely

Beyond CVE-2026-8461, three principles keep FFmpeg safe. First, keep applying the latest fixes. Most vulnerabilities are closed by updates. As here, "even 8.1.2 leaves a new flaw" can happen, so don't treat one update as done—check the FFmpeg security page and vendor advisories regularly. Second, don't process files of unknown origin. FFmpeg's weak spots concentrate in video/subtitle parsing, so simply not opening untrusted files greatly lowers risk. Third, isolate conversion when you use it in a service. When converting user-uploaded files, restrict formats and process them in an environment separated from the main system (a sandbox) so that even a successful exploit is contained.

If you cannot update immediately, stopgaps such as not processing video/subtitle files of unknown origin and restricting formats before converting user uploads buy time. But the real fix is updating. If you cannot fully account for which software and versions your service uses, auditing your open-source (OSS) dependencies together is the fast path. A mechanism that surfaces dangerous versions from a list of the components you use—like our OSS vulnerability scanner—makes it easier to catch "we were unknowingly running an old component" situations.

Why do FFmpeg vulnerabilities keep piling up?

FFmpeg vulnerability reports have risen noticeably in 2026. One reason is the spread of AI-assisted vulnerability discovery mentioned above. In June 2026 an AI agent was reported to have automatically analyzed FFmpeg's roughly 1.5 million lines of code and found 21 unknown vulnerabilities (zero-days), each said to come with a reproducible attack procedure—big news in the security field.

FFmpeg underpins countless services, yet is maintained by a small group of volunteers. As a few people guard an enormous codebase, we have entered an era where AI unearths weaknesses one after another. All four July flaws and June's CVE-2026-8461 can be seen as flaws surfaced amid this trend of "foundational software's holes being dug up in bulk." This is not "FFmpeg is dangerous," but a shift across all software: the more widely used the foundation, the more its weak spots come to light. For anyone who handles video, frequent updating matters more than ever.

Frequently asked questions

Q. I updated to 8.1.2 in June. Am I safe now?

A. June's CVE-2026-8461 is closed by 8.1.2. But the four July flaws (CVE-2026-64830, 64831, 64832, 64835) are unfixed even in 8.1.2, so you need to apply an additional latest fix (your OS security update or a newer build). Don't treat one update as done—check your vendor's advisories.

Q. Is FFmpeg safe? Can I keep using it?

A. FFmpeg itself is a long-used, trustworthy free tool—not one to avoid. The two conditions for safe use are "keep applying the latest fixes" and "don't process video/subtitle files of unknown origin."

Q. Can I be attacked through the MP4 or MOV videos I normally use?

A. These flaws are in specific processing—subtitles (VobSub), HEVC hardware playback, and MagicYUV. They don't necessarily trigger merely from playing or converting ordinary MP4 or MOV files. That said, attackers can disguise containers and extensions, so it is safest not to handle files of unknown origin.

Q. I'm worried I may already have been attacked.

A. None of these flaws had confirmed real-world exploitation at publication. Even so, attempts become more likely after disclosure, so update sooner rather than later.

Summary

The answer to "is FFmpeg safe" is "safe as long as you apply the latest fixes and handle only trusted files; dangerous if left old." In 2026, AI-assisted discovery has accelerated and FFmpeg reports have piled up, but that is a shift where widely used foundations surface more weak spots—not a reason to avoid the software itself.

The top priority right now is the four flaws published July 23 (CVE-2026-64830, 64831, 64832, 64835). All are severity 8.8 and remain in June's fixed release 8.1.2. Also update for June's CVE-2026-8461 (fixed in 8.1.2) if you haven't. FFmpeg runs as a component under many apps and services, so "I don't use it" can still mean indirect exposure. The surest way to keep it safe long-term is to make updating your dependencies a habit—especially if you run anything that handles video.

References

• ▸ NVD - CVE-2026-64830 (VobSub subtitle demuxer out-of-bounds write, CVSS 8.8)
• ▸ NVD - CVE-2026-64831 (Vulkan HEVC decoder stack overflow, CVSS 8.8)
• ▸ NVD - CVE-2026-64832 (NVIDIA NVDEC decoder double free, CVSS 8.8)
• ▸ NVD - CVE-2026-64835 (ADX audio decoder out-of-bounds access, CVSS 8.8)
• ▸ FFmpeg - fix pull request #23657 (VobSub)
• ▸ FFmpeg - fix pull request #23659 (ADX audio)
• ▸ NVD - CVE-2026-8461 (MagicYUV decoder out-of-bounds write, CVSS 8.8)
• ▸ FFmpeg - fix pull request #23159 (MagicYUV)
• ▸ FFmpeg - Security
• ▸ FFmpeg - Download
• ▸ The Hacker News - AI Agent Uncovers 21 Zero-Days in FFmpeg (June 2026)
• ▸ Cyber Security News - 21 0-Day Vulnerabilities in FFmpeg (June 2026)