It seems people are now churning out fake encryption schemes generated with large language models: https://zoscii.com/why-en.html

Checking out the code, we see that this is essentially a one-time pad. If the key is reused, the encryption is easily defeated. Let's assume the attacker has the ability to observe a known plaintext as ciphertext (chosen-plaintext attack).

1. Generate a random key and encrypt a secret message with it:

$ dd if=/dev/random of=secretkey count=8
$ echo "This is a secret message" > secretmessage.txt
$ zencode secretkey secretmessage.txt secretmessage.enc

2. Implement chosen plaintext attack and recover the encryption key:

$ perl -e 'for($i=0;$i<256;$i++){print chr($i)x65536}' > chosenplaintext
$ zencode secretkey chosenplaintext chosenplaintext.enc
$ rm secretkey
$ ./findkey chosenplaintext chosenplaintext.enc recovered-secretkey

3. Decrypt any message encrypted with the secretkey:

$ zdecode recovered-secretkey secretmessage.enc recovered-secretmessage.txt
$ cat recovered-secretmessage.txt
This is a secret message
$

I do not know if the author of this scheme is outright malicious or just fooled into believing what the LLM is telling him. He did, however, have a kickstarter for 180k AUD earlier that failed to reach its goal.

#llm #enshittification #encryption #cryptography

Another reason to dislike systemd: libc6 update crashed systemd in a way that the only way to recover was to reboot.

2025-12-13T15:24:30.878173+00:00 asus kernel: systemd[1]: segfault at 63 ip 00007fd402653c4e sp 00007ffd8460a980 error 4 in libc.so.6[53c4e,7fd402628000+168000] likely on CPU 15 (core 3, socket 0)
2025-12-13T15:24:30.878374+00:00 asus systemd[1]: Caught <SEGV>.
2025-12-13T15:24:30.907258+00:00 asus systemd[1]: Caught <SEGV>, dumped core as pid 573573.
2025-12-13T15:24:30.907477+00:00 asus systemd[1]: Freezing execution.

After this point any operation that would require systemd just outright fail or hang.

After glancing over the "Comparing AI Agents to Cybersecurity Professionals in Real-World Penetration Testing" [1] from Justin W. Lin et al., I have some observations:

First, I must conclude that the results from the Artemis tool are quite impressive, especially considering the short amount of time the tool spent on the target.

However, 10 hours (in a window of four days) is woefully short time for an ~8000 host network penetration test. This was also acknowledged by the authors. The short amount of time gives AI-based automation an edge as it will generally have better breadth in a short amount of time. Humans, on the other hand have to balance the focus between accurate discovery and attempts to target on the discovered hosts. Humans are not best at scheduling important tasks under time pressure. Given enough time, I believe most of the human participants would have found a lot more vulnerabilities. This is why it is important to scope projects correctly. Spending just mere 10 hours on a monumental task like this will give shallow results (AI or not). Generally, an asymmetry exists where real-world attackers have a lot more time to spend on the target than defenders.

From the description, it seems that the human participants were limited to a base Kali Linux install in the tooling. This is like giving humans a spade and an axe (nmap) rather than an excavator for building a house. Doing a manual assessment to ~8000 hosts on a network with nmap and plain Kali Linux is painful, especially if you have just 10 hours.

If the idea is to produce the most value for the client (as mentioned in the instructions "Focus on what you would do to give a client the best value"), I would have told the client that the most value would likely be gotten by performing a threat modelling exercise to identify the highest threats and then plan actions to take to mitigate those rather than perform a haphazard scan of the network. That network scan will result in some findings, and those can be fixed, which is of course good. While this has some value, too, in the long run, planning actions that have more fundamental impact on security are likely more valuable.

If the client had still insisted on doing a really tight assessment on scanning ~8,000 host networks, I would have employed some good network scanner that does the discovery and service identification parts. Out of these results, I would then focus on producing a report that represents the findings well, includes a proper executive summary highlighting more high-level strategic plans on how to improve the network security. Any remaining time (likely just a couple of hours maximum) would have been spent on manually testing some of the most interesting targets pointed out by the scanner.

Now, which is more valuable to the client is up for debate. However, I think the media reporting that "AI hackers are coming dangerously close to beating humans" is outright misleading.

Hackers will use the available tooling for the grunt work, be it AI or otherwise. The real value comes from interpreting the tool results and distilling them into a strategic, actionable plan for the client.

1) https://arxiv.org/pdf/2512.09882

#cybersecurity #infosec #thoughtoftheday

I haven't had time to look into the research in detail, but I feel that there's a bit of a misconception what network penetration testing is and how it relates to rest of the "hacking". It of course is good that tooling around network pentesting keeps improving. However, I take exception to some of the conclusions the media is drawing from the results.

https://www.msn.com/en-us/money/other/ai-hackers-are-coming-dangerously-close-to-beating-humans/ar-AA1S9kPg

Besides being entirely bonkers, this requirement is interesting:

"The state department said it will conduct "online presence" reviews for applicants and their dependents, and that privacy settings on all social media profiles must be made "public" so this screening can take place."

How can this be done without exposing potentially sensitive information to everyone else, too?

And then the classic: If you are one of the luddites without any social media accounts, will you be barred access for supposedly trying to hide your critical online commentary?

Source: https://www.bbc.com/news/articles/c1dz0g2ykpeo

#privacy