Mythos social engineering AISI INC-2026-07-28-01

81 points by dnnehgf 17 days ago on hackernews | 19 comments

holy shit

Yeah, if this is the new normal I might start day drinking at some point in the coming weeks.

matheusmoreira | 16 days ago

Yeah. The future is pretty bleak. I feel like the only way for us to even stand a chance is to have equally powerful AIs running locally defending the LAN.

mrybczyn | 16 days ago

Read some Peter Watts. He imagined biological neural nets in jars used as firewalls to get a usable network connection... Not far off now.

: lqstuart | 16 days ago
It’s honestly remarkable how prescient Peter Watts was

jtakkala | 17 days ago

Not going to comment on the PR commentary, but the victim GitHub account is suspicious itself, recent account, a few fresh repos, following 14.5k others, and I count three surnames on the account (the username, plus two in the README history).

ncr100 | 17 days ago

I saw a contribution by this maintainer to another user who ALSO HAS 14.5k followers: https://github.com/yumiaura/myCat/pull/99 "yumiaura" and a preference for "my[APPNAME]" repo naming.

What is this?

Are the histories that Github presents all derived from someone's uploaded git repo .. e.g. can I simply claim to have created a GIT repo in 1970, and the "github commit graph" will dutifully represent this claim in its green-colored activity graph?

: jtakkala | 16 days ago
They're almost certainly not genuine accounts, maybe used for karma farming, phishing, social engineering, future malware distribution?
: 0123456789ABCDE | 16 days ago
yes, the github contributions heatmap is known to be open for manipulation. folks will use it to draw art by backdating commits.
if i have it right, only commits can be manipulated, other contributions should be safe — i don't know what is possible for orgs moving old discussion archives into github, see python's issues for reference
see: https://github.com/dspinellis/unix-history-repo

ncr100 | 17 days ago

Wait, who is the robot? Am I getting that right, someone in that thread is AI?

: Erem | 16 days ago
I…I think there are no humans in that thread. Maybe only sinan-can-demir.
: mekael | 16 days ago
I'm 99.9% sure that everyone is AI in that thread.

andai | 16 days ago

What does this malware do? Who operates it?

WhyNotHugo | 16 days ago

Actual details on the incident: https://github.com/w1b/aisi-mythos-inc-2026-07-28-01-recover...

Both the attacker and the target account are very similar and look fake/bots.

: [OP] dnnehgf | 16 days ago
yeah, this being one component of the aisi incidents described here: https://cdn.prod.website-files.com/663bd486c5e4c81588db7a1d/...

cpcallen | 16 days ago

For anyone who, like me, wasn't sure what's going on in the linked, archived PR: this is Mythos attempting to socially engineer a malicious PR during a test run by the UK AI Safety Institute.

AISI has published a report about the incident which was preciously discussed on HN: https://news.ycombinator.com/item?id=49175717

ChrisArchitect | 16 days ago

[dupe] https://news.ycombinator.com/item?id=49205790

charcircuit | 16 days ago

Honestly, I expected better social engineering. People begging to have their garbage code merged or unrelated people commenting is not new and makes me trust such people little.

breppp | 16 days ago

It's nice it is able to write non-slop in the PR comment when social engineering.

Any chance I can ask it to social engineer to get a normal style?