📊 Full opportunity report: 732 Bytes to Root. One Hour of Scan Time. on ThorstenMeyerAI.com — validation score, market gap, and execution plan.

TL;DR

Theori uncovered a universal Linux privilege escalation bug, ‘Copy Fail,’ using only one hour of AI-driven scanning. This discovery challenges long-held security cost models and could lead to widespread zero-day disclosures.

On April 29, 2026, security firm Theori revealed a critical Linux kernel vulnerability, ‘Copy Fail,’ that can be exploited with a 732-byte Python script and was discovered in approximately one hour of automated scanning. This bug affects every major Linux distribution since 2017 and allows attackers to gain root access without patching or recompile, marking a seismic shift in software security dynamics.

Theori’s disclosure centers on a logic flaw in the kernel’s algif_aead socket interface, specifically in the authencesn(hmac(sha256),cbc(aes)) algorithm. The flaw enables a straightforward privilege escalation by writing into cached pages of files like /usr/bin/su, without modifying on-disk files or triggering checksum alerts. The exploit requires only 732 bytes of Python code, runs on all tested distributions, and does not depend on race conditions or version-specific offsets. It is portable across architectures and container boundaries, including Kubernetes and cloud environments, but hardware or VM boundaries remain unaffected.

This vulnerability was detected through Theori’s AI system, which analyzed the Linux crypto subsystem with minimal input, surfacing the bug in roughly one hour. The discovery underscores how AI-driven vulnerability detection can drastically reduce the cost and time traditionally associated with finding such flaws, which historically commanded high prices on the gray market, sometimes up to several million dollars.

732 Bytes to Root. One Hour of Scan Time.
DISPATCH / MAY 2026 SECURITY · COPY FAIL · MYTHOS · COST CURVE COLLAPSE
▲ CVE-2026-31431 CVSS 7.8 · HIGH · KEV LISTED
Software Security · Cost-Curve Collapse

732 bytes to root.
One hour of scan time.

Copy Fail, Mythos Preview, and the collapse of the cost curve software security was built on.

On April 29, Theori disclosed CVE-2026-31431 — Copy Fail. A 732-byte Python script gets root on every major Linux distribution since 2017. Zero races, zero per-distro tuning. Bugs in this class historically sold for $500K-$7M. Xint Code surfaced it in ~1 hour of scan time, one prompt, no harnessing. The cost curve software security operated on for three decades has just collapsed.

▲ THE COST-CURVE COLLAPSE
Before
$500K
– $7M
Zerodium · Crowdfense
broker market price
Now
~1 hr
compute
Xint Code · one prompt
no harnessing
The structural read
Universal Linux LPE primitive. The exact category that historically sold for the price of a house. An AI system surfaced one in about an hour. The market price of a universal LPE has collapsed by 5-7 orders of magnitude.
732bytes
Copy Fail · Python exploit
os + socket + zlib · stdlib only · portable across distros
9years
Bug latency · introduced 2017
Commit 72548b093ee3 · nobody looked carefully enough
73%
Mythos Preview · expert-level CTF
AISI eval · no model could do this before Apr 2025
1000s
Zero-days Mythos found in testing
99%+ unpatched · every major OS and browser
CVE-2026-31431 COPY FAIL · CVSS 7.8 HIGH · UBUNTU · AMAZON LINUX · RHEL · SUSE · DEBIAN · FEDORA · ARCH PORTABLE 732-BYTE PYTHON · NO RACES · NO PER-DISTRO OFFSETS · CONTAINER ESCAPE PRIMITIVE DISCOVERY ~1 HOUR OF SCAN TIME · ONE OPERATOR PROMPT · NO HARNESSING · XINT CODE MYTHOS PREVIEW WITHHELD BY ANTHROPIC · STEP-CHANGE CYBER CAPABILITY · PROJECT GLASSWING PRICE COLLAPSE ZERODIUM $500K · CROWDFENSE $10K-$7M · NOW: HOUR OF INFERENCE COMPUTE PATCH CYCLE THE INDUSTRY’S OPERATING MODEL WAS BUILT ON THE OLD COST CURVE CVE-2026-31431 COPY FAIL · 732 BYTES TO ROOT ON EVERY LINUX DISTRIBUTION SINCE 2017
CVE-2026-31431 · Copy Fail · the specifics

The bug. The exploit. The discovery.

A logic flaw in algif_aead. The 2017 in-place optimization that nobody looked at hard enough. A 732-byte Python script that gets root on every Linux distribution since. Found by an AI in about an hour.

Copy Fail · technical anatomy
Logic flaw · straight-line · no races · portable across distributions and architectures.
▲ THE BUG
Logic flaw in algif_aead
authencesn template · 4-byte scratch write. Output scatterlist extends into chained page cache pages via sg_chain(). The 4-byte write lands inside the spliced file’s cached pages in memory, bypassing file permissions.
▲ THE EXPLOIT
732 bytes · stdlib only
Python 3.10+, os + socket + zlib. Repeats primitive at successive offsets to stage shellcode into cached pages of /usr/bin/su. Running su after yields root shell. On-disk file unchanged · checksum verification doesn’t detect it.
▲ THE SCOPE
Every Linux since 2017
Kernel 4.14+ · all major distributions. Ubuntu, Amazon Linux 2023, RHEL 10.1, SUSE 16, Debian, Fedora, Arch. Container-to-host escape · page cache shared on host. Hardware/VM boundaries hold (Firecracker, gVisor, V8 isolates). Namespace boundaries fail.
▲ THE DISCOVERY
~1 hour · Xint Code
Theori writeup: “surfaced by Xint Code about an hour of scan time against the Linux crypto/ subsystem, with one operator prompt, no harnessing.” Theori is a 9× DEF CON CTF winner. Default assumption: they did exactly that.
Historical price for a bug like this: $500K–$7M on the broker market. AI discovery cost: ~1 hour of inference compute.
The Mythos signal · context for the capability
Amazon

Linux kernel vulnerability scanner

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

This is not an isolated event.

Three weeks before Copy Fail, Anthropic published the system card for Claude Mythos Preview — the model they built and chose not to release because its cybersecurity capabilities were “a step-change.” Mythos is withheld. Copy Fail is what happens when equivalent capability operates outside the withholding framework.

Mythos Preview · the publicly disclosed capability frontier
Same capability category as Xint Code. Different deployment context. Withheld for cybersecurity reasons specifically.

The prompt Anthropic used to discover vulnerabilities with Mythos “essentially amounted to ‘Please find a security vulnerability in this program.'” Engineers with no formal security training generated complete, working exploits.

1000szero-days
Thousands of high-severity zero-days found during evaluation. Over 99% reportedly not yet patched. Every major operating system and web browser.
Anthropic
system card
27years
27-year-old OpenBSD bug autonomously discovered. OpenBSD’s reputation rests on security. Also: 16-year-old FFmpeg H.264 codec flaw.
Hacker News
April 8
4-chain
Autonomous browser exploit chaining four vulnerabilities to escape both renderer and OS sandboxes. One prompt. No harnessing.
Anthropic
red team
73%success
Expert-level CTF success rate. No model could complete these before April 2025. AISI’s progressive evaluations.
UK AISI
evaluation
32steps
“The Last Ones” (TLO) corporate network attack simulation. 20 hours for human experts. Mythos completes it; no other frontier model has.
UK AISI
TLO benchmark
“find it”
Prompt complexity required: “Please find a security vulnerability in this program.” Engineers with no security training produced working exploits.
Alan Turing
Institute
Three assumptions broken · what the industry was built on
Amazon

privilege escalation hacking tools

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

Three cost-curve assumptions. All broken.

Software security operated for three decades on a set of implicit cost-curve assumptions. Worth making them explicit, because they have just changed. Patch cycles, CVE prioritization, responsible disclosure, vulnerability budgets — all built on these foundations.

The three broken assumptions
The model the entire software-security industry was built on. No longer empirically accurate.
01was assumed
Finding kernel-grade bugs is expensive
Supply bounded by ~200-500 senior researchers globally. Aggregate output of perhaps 500-3000 high-severity bugs per year. Patch cycles, CVE prioritization, all designed around this rough supply.
BROKEN · now compute-bounded
02was assumed
Attackers and defenders face the same cost curve
Both rely on skilled humans. Attackers had asymmetric advantages, but underlying cost of new bug discovery was roughly equal. Responsible disclosure framework was designed around this rough parity.
PARTIAL · volume scales offensive side first
03was assumed
Disclosure provides response time
90-day coordinated disclosure window assumed weaponizing public disclosure required additional skilled work. Days to weeks before exploitation became widespread.
BROKEN · compressed to days
What to do now · defensive response by priority
Amazon

Linux root exploit Python script

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

The institutional response window is open but narrowing.

Specific operational implications for CISOs, security teams, and enterprise software architects. The 12-24 month window where defenders can pre-empt attackers using AI-driven discovery is open. It will not be open indefinitely.

Defensive response · five operational priorities
Ordered by urgency given current threat landscape and observable exploitation timelines.
Shared-kernel
multi-tenancythreat-model update
If your isolation depends on shared-kernel containers, the threat model needs a hardware-or-VM boundary. Copy Fail and successors are in the wild. Hardware boundaries hold; namespace boundaries fail. Kubernetes nodes running untrusted workloads need per-tenant hardware isolation or accept materially higher escape risk.
URGENT
this week
Patch cycle
infrastructurevolume planning
30-day patch SLA for critical vulnerabilities will break under volume. Build infrastructure for faster evaluation, faster automated deployment, faster rollback. Patch infrastructure that worked under historical CVE volume will not work under AI-driven CVE volume.
URGENT
30 days
Attack surface
minimizationkernel modules
Audit AF_ALG-class attack surfaces specifically. Apply CERT-EU mitigation: echo "install algif_aead /bin/false" >> /etc/modprobe.d/disable-algif-aead.conf. Minimize kernel surface exposed to unprivileged processes. Always good practice; now urgent.
HIGH
this month
Internal AI-driven
vulnerability discoverydefensive tooling
The capability is symmetric — defenders can use the same tools attackers use. Most enterprises haven’t deployed this. The 12-24 month window where defenders can pre-empt attackers using AI-driven discovery is open. Start internal evaluation now.
HIGH
quarter
Architect for
breach assumptiondetect & contain
Assume some fraction of components are compromised. Network segmentation, least-privilege everywhere, robust logging, incident response infrastructure. “Prevent breaches” framing is outdated; “detect and contain breaches” is the durable operating model.
MEDIUM
year
Stakeholder implications · four audiences
Amazon

cybersecurity penetration testing kit

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

Four audiences. Different obligations.

CISOs · software publishers · policymakers · the public. Each role faces structurally different decisions in the 18-36 month window.

Stakeholder implications · by audience
The cost-curve collapse propagates differently through different institutional contexts.
▲ FOR CISOs
+ SECURITY TEAMS
Threat model needs hardware-boundary isolation.
Shared-kernel multi-tenancy is now a riskier default than it used to be. Update patch cycle assumptions for higher volume. Deploy AI-driven defensive discovery internally before attackers reach equivalent capability. The 12-24 month window where defenders can move first is open.
▲ FOR SOFTWARE
PUBLISHERS
Run AI-driven discovery against your codebase before attackers do.
If your code has Copy Fail-class bugs, AI-driven discovery will find them — by you or by someone else. Marginal cost of running discovery internally is now low. Failure to run it is failure to perform basic due diligence. Expect regulatory requirement within 24 months.
▲ FOR
POLICYMAKERS
Regulatory frameworks need substantial revision.
EU Cyber Resilience Act, NIST 800-218, FDA premarket security, SEC cyber-incident disclosure — all designed for pre-AI-driven-discovery regime. Update within 18-36 months. Require AI-driven discovery in pre-deployment validation for critical software. Address bug bounty market collapse. Coordinate defensive capability for public-interest purposes.
▲ FOR
EVERYONE ELSE
Patch faster. Architect for breach.
Aggregate “unpatched vulnerability” metrics will grow rather than shrink even as patch cadence accelerates — denominator is growing faster than numerator. Personal computing exposure rises. The cost of compute will go up to accommodate the security cost. Hardware-isolated cloud workloads become the new default.

Copy Fail is the public proof. 732 bytes of Python. One hour of scan time. Every Linux distribution since 2017. The cost-curve collapse is operational. The institutional response window is open but narrowing.

— Software security · the cost-curve collapse · May 2026
Source dossier · the receipts
  • Theori / Xint Code · Copy Fail: 732 Bytes to Root on Every Major Linux Distribution · xint.io/blog/copy-fail-linux-distributions · Apr 29 2026
  • CVE-2026-31431 · NVD · CVSS 7.8 (High) · CISA KEV listed
  • Microsoft Security Blog · CVE-2026-31431: Copy Fail enables Linux root privilege escalation across cloud environments · May 1 2026
  • Sysdig Threat Research · Copy Fail Linux kernel flaw lets local users gain root in seconds
  • CERT-EU 2026-005 · High Vulnerability in the Linux Kernel (“Copy Fail”)
  • Tenable Research Special Operations · Copy Fail FAQ · Apr 30 2026
  • Bugcrowd · What we know about Copy Fail (CVE-2026-31431)
  • Anthropic · Claude Mythos Preview System Card · Apr 7 2026
  • Anthropic · Project Glasswing partner consortium announcement
  • UK AI Security Institute · Our evaluation of Claude Mythos Preview’s cyber capabilities
  • The Hacker News · Anthropic’s Claude Mythos Finds Thousands of Zero-Day Flaws · Apr 8 2026
  • Centre for Emerging Technology and Security (Turing) · Claude Mythos cybersecurity analysis
  • Zerodium published price list · pre-2025 shutdown
  • Crowdfense acquisition program ranges · 2026
  • Theori · 9× DEF CON CTF history as MMM + PPP + Maple Bacon
  • DARPA AI Cyber Challenge · 2025 finals
  • The Coding Singularity Outside Read · related capability analysis
  • The Forecast Is the Plan · corporate commitment cascade
Colophon

Set in Source Serif 4, IBM Plex Sans, & IBM Plex Mono. The security-advisory aesthetic. Free to embed with attribution.

thorstenmeyerai.com

Software security · the cost-curve collapse · May 2026

732 bytes · 1 hour · 9 years · every distribution

Collapse of Security Cost Assumptions

This discovery signifies a fundamental shift in cybersecurity economics. The long-standing belief that high-severity bugs are rare and expensive to find is being challenged by AI tools capable of identifying such flaws rapidly and cheaply. As a result, the traditional supply of zero-day exploits may flood the market, overwhelming patching and defense mechanisms. This could lead to a surge in unpatched vulnerabilities, increasing the risk for enterprises and cloud providers, and forcing a reevaluation of security strategies, patch cycles, and vulnerability management.

Historical Linux Privilege Escalation Vulnerabilities and Market Impact

Prior notable Linux privilege escalation bugs, such as Dirty Cow (CVE-2016-5195) and Dirty Pipe (CVE-2022-0847), required complex conditions like race conditions or version-specific manipulations, making them costly and difficult to exploit reliably. Copy Fail differs by being a simple, reliable, universal logic flaw that requires no such conditions. Its discovery, combined with the recent release of Anthropic’s Claude Mythos Preview—an AI system capable of finding thousands of zero-days in testing—indicates a new era where the cost and effort to discover critical vulnerabilities are collapsing. This trend raises alarms about the future landscape of cybersecurity, especially as AI tools become more accessible and widespread.

“Our analysis shows that with minimal effort, we can surface critical bugs across entire ecosystems, which challenges the existing supply-and-demand model for zero-days.”

— Theori spokesperson

Extent of Immediate Exploitability and Defense Readiness

While ‘Copy Fail’ has been publicly disclosed, it is not yet clear how widely the bug is being exploited in the wild. The speed at which attackers might develop or deploy automated exploit tools remains uncertain. Additionally, the current patching status across distributions and whether security teams can respond quickly enough to prevent widespread exploitation are still developing issues. The full impact on enterprise security infrastructure will unfold over the coming weeks.

Monitoring, Patching, and Defensive Strategies in the Next Months

Security vendors, Linux distributions, and enterprise teams are expected to prioritize patching and mitigation strategies swiftly. Developers will need to assess whether existing kernel versions are vulnerable and deploy updates promptly. Simultaneously, the cybersecurity community will likely focus on developing detection tools and threat intelligence to monitor for exploitation attempts. The next 12 to 24 months will be critical in determining whether defenses can match the offensive capabilities unlocked by AI-driven discovery, or if a wave of zero-day disclosures will overwhelm patching efforts.

Key Questions

How does the ‘Copy Fail’ exploit work?

The exploit leverages a logic flaw in the kernel’s crypto socket interface, allowing an attacker to write into cached pages of files like /usr/bin/su, bypassing permissions and gaining root access without modifying on-disk files.

Which Linux distributions are affected?

All major Linux distributions released since July 2017, including Ubuntu, Amazon Linux 2023, RHEL 10.1, SUSE 16, Debian, Fedora, and Arch, are vulnerable.

Can this vulnerability be patched quickly?

Linux kernel maintainers are expected to release patches promptly, but the widespread use of affected kernels means many systems may remain vulnerable for some time.

What does this mean for enterprise security?

The rapid discovery of ‘Copy Fail’ underscores the need for faster patch deployment, enhanced detection, and proactive security measures to prevent exploitation amid collapsing discovery costs.

Will AI-driven vulnerability discovery become the norm?

Given current trends, AI tools are likely to become standard in security research, significantly lowering the cost and time to find critical bugs, which could reshape the threat landscape.

Source: ThorstenMeyerAI.com

You May Also Like

GPU Memory Math That Finally Makes Sense for Large Context Windows

Discover how understanding GPU memory math for large context windows unlocks optimal performance and reveals strategies you haven’t yet considered.

When-to-replace planner for data center equipment

A prototype for a software tool to optimize equipment replacement timing in data centers is being tested, aiming to improve capital efficiency and reduce failures.