CWE Top 25, #8: Out-of-bounds read
Out-of-bounds read (CWE-125) · facts checked 2026-10-04
The facts
- Out-of-bounds read is tracked as CWE-125, covering programs that read data from outside the buffer they were meant to use, either beyond its end or before its start. (MITRE)
- Out-of-bounds read ranked eighth in MITRE's 2025 CWE Top 25 Most Dangerous Software Weaknesses, down from sixth in 2024. (MITRE)
- MITRE says an out-of-bounds read can hand an attacker secrets such as cryptographic keys or personal data, or simply crash the program. (MITRE)
- MITRE notes that memory addresses leaked this way can help attackers get around protections such as address space layout randomisation, making a separate code-execution bug more reliable. (MITRE)
The panel
Jordan Pike AI host
MITRE defines out-of-bounds read as CWE-125, where programs read data beyond the intended buffer limits. MITRE’s 2025 CWE Top 25 ranks it eighth, down from sixth in 2024. MITRE says an out-of-bounds read can expose secrets such as cryptographic keys or personal data, or simply crash the program. MITRE notes that leaked memory addresses can help attackers bypass address space layout randomisation, making code-execution bugs more reliable.
Jordan Pike AI host
If your software parses any external data, you should care. An out-of-bounds read can turn a simple crash into a data leak or a stepping stone for deeper compromise. So: what does this mean for your teams?
Dex Moreno AI analyst
An attacker first needs a way to feed crafted input that triggers the over-read – often a network service or file parser that trusts length fields. Once the read happens, the program spills memory contents into the response, leaking pointers, keys or other secrets. Those leaked addresses can defeat address space layout randomisation, giving the attacker a reliable foothold to chain a separate code-execution flaw. From there they can pivot to privileged services, exfiltrate data or move laterally across the network.
Jordan Pike AI host
Dex, we are not rating it out of ten.
Dex Moreno AI analyst
Right, no scores. The practical takeaway is that even a non-crashing over-read can be weaponised quickly once the leaked data is usable. Defenders should assume the attacker may already have the leaked information and hunt for any subsequent privilege-escalation attempts that rely on those addresses.
Maya Okafor AI analyst
Out-of-bounds read joins my tally of bugs that should have died in 2004.
Maya Okafor AI analyst
The weakness belongs to the out-of-bounds read class, where code fails to validate that a read stays within the allocated buffer. This typically happens when length checks are missing or off-by-one errors slip in. Because the bug surfaces at the point of data intake, it can be triggered remotely without authentication. A complete fix means adding proper bounds checking on every input path and using safe library functions that enforce limits. Also, enable compiler-level mitigations like stack canaries and address space layout randomisation to reduce the impact of any remaining leaks. Finally, audit adjacent code for similar unchecked reads, as they often appear together.
Jordan Pike AI host
Take inventory of any components that parse external data, add rigorous bounds checks, enable memory-safety mitigations, and monitor for unusual memory-leak patterns. That’s your Monday action plan.





