This week brings up a pretty solid variety of issues. Starting off with some cookie smuggling (and other cookie attacks) which presents some interesting research I hadn't really looked for before that has some potential. Then an AI alignment evasion to leak training data. Not the most interesting attack but it appears to open up some other ideas for further research. A MacOS desktop issue (for a $30k bounty), and some home assistant issues.
This week kicks off with a a V8 misoptimization leading to out-of-bounds access, an unprotected MSR in Microsoft's Hypervisor allowing corruption of Hypervisor code. We also take a quick look at a 2021 CVE with an integer underflow leading to an overflow in the Windows Kernel low-fragmentation heap, and finally an interesting information leak due to the kernel not clearing a sensitive register.
This week we've got a few relatively simple bugs to talk about along with a discussion about auditing and manually analysis for vulnerabilities.
Last week we brought you several Windows bugs, this week we are talking Linux kernel vulnerabilities and exploitation. We start off looking at a weird but cool CPU bug, Reptar, then we get into nftables, io_uring, and talk about a newer mitigations hitting Linux 6.6 that randomizes the caches allocations end up in.
This week has an interesting mix of issues, starting with a pretty standard template inject. Then we get into a Windows desktop issue, a TOCTOU in how the Mark-of-the-Web would be applied to file extracted from an archive, a privilege escalation from a Chrome extension, and a bit of a different spin on what you could do with a prompt injection.
We've got a few Windows bugs this week, but first a fun off-by-one null-byte write. Then we jump into a containerized registry escape, a browser escape with a very simple bug buried deep in the browser, and a kernel bug.
Just a few issues this week, a Mastodon normalization issue leading to the potential to impersonate another account. Then we have a more complex chain starting again with a normalization leading to a fairly interesting request smuggling (CL.0 via malformed content-type header) and cache poisoning to leak credentials. Finally a crypto issue with a signature not actually being a signature.
As memory tagging (MTE) finally comes to a consumer device, we talk about how it may impact vulnerability research and exploit development going forward. Then we get into a few vulnerabilities including a DNS response parsing bug on the Wii U, an Adobe Acrobat bug that was exploited by a North Korean APT, and a CPU bug (iTLB Multihit).
Kicking off the week with a bit of Pwn2Own drama, then taking a look at an OAuth attack against Grammarly and a couple other sites, a fun little polyglot file based attack, and Citrix Bleed, a snprintf information disclosure vulnerability on the web.
Diving right into some binary exploitation issues this week. Starting wtih a look at a rare sort of curl vulnerability where a malicious server could compromise a curl user. Then we take a look at a pretty straight-forward type confusion in Windows kernel code, and an integer underflow in Safari with some questionable exploitation. Ending the episode with some thoughts on how impactful grsecurity's "constify" mitigation could be.