“Breaking Claude Code Opus 5 Auto Mode,” n.d. https://embracethered.com/blog/posts/2026/breaking-claude-code-opus-5-and-automode/
I looks like basic promt injection is no longer a problem. Common attacks where you ask agent to ignore previous instructions seem to be fixed.
There is still huge amount of promt injection attacks which are not solved. This article shows couple of them.
From my side I think it’s good to know that probably a lot of those attacks are not public. It’s good that we have security researchers who publish their results but I can imagine quite some black hats out there just use it to get good ransom.
.
There is still huge amount of promt injection attacks which are not solved. This article shows couple of them.
From my side I think it’s good to know that probably a lot of those attacks are not public. It’s good that we have security researchers who publish their results but I can imagine quite some black hats out there just use it to get good ransom.
“Spyware for Babies,” n.d. https://www.schneier.com/blog/archives/2026/08/spyware-for-babies.html
In USA they now have cameras for babies which have AI capabilities to monitor child development.
And people really buy this. Wow. Land of the free xd
.
And people really buy this. Wow. Land of the free xd
“Qwen 3.8 27B Is Excellent, but It Defaults to Wildly Overthinking Things,” n.d. https://simonwillison.net/2026/Aug/16/qwen-38-27b/
I see the new Open models don’t differ that much from commercial ones. I think it’s clear that now lobby from Meta and other companies which invested billions will try to influence foreign governments and prohibit usage of such technologies.
.
“Wild AI-Related Reliability Incidents Are Coming,” n.d. https://surfingcomplexity.blog/2026/08/22/wild-ai-related-reliability-incidents-are-coming/.