Flux
Self-generated prompt injections in compaction summaries
IA Programmation

Self-generated prompt injections in compaction summaries

Self-generated prompt injections in compaction summaries In Our framework for reporting model misalignment OpenAI provide "six reports on unexpected or concerning model behavior we’ve observed in the last six months". This one here is my favorite: they caught some of their models in training deliberately subverting themselves in their compaction prompts. Compaction is the process agent systems use when they are running out of tokens in their context window, so they summarize everything that has…

Simon Willison's Weblog
PolinRider Spreads Through Compromised GitHub Accounts and Packagist
Cybersécurité Programmation

PolinRider Spreads Through Compromised GitHub Accounts and Packagist

Socket researchers identified malicious code in the dev-main version of visanduma/nova-two-factor, a Packagist package with more than 700,000 cumulative downloads, as the PolinRider campaign continues to spread through compromised developer accounts and Git repositories. The Socket Threat Research Team continues to track malicious activity associated with PolinRider. In our July research post, we provided initial technical details about this persistent campaign, which distributes malware across…

Socket
datasette 1.0a40
IA Programmation

datasette 1.0a40

Release: datasette 1.0a40 Same security fix as 0.65.5, plus some neat new features and bug fixes: Plugins can now launch and manage background tasks using the new datasette.add_background_task() method. Thanks, Alex Garcia. I've migrated Datasette to httpx2 for features like the internal datasette.client.get() method. A whole lot of bug fixes, many of them stemming from a recent effort to triage issues for a 1.0 stable release. Tags: security, datasette

Simon Willison's Weblog
Modernize with htmx
Programmation Web

Modernize with htmx

Welcome to the September issue of PHP Architect. Stop everything! Did you know that at this very minute, Tek 27 Call For Papers is LIVE?! It gets better… You can also purchase early-bird tickets to Tek 27 right now!! Breathe, take a minute, let yourself process that amazing info. Now that you’ve let that sink […] The post Modernize with htmx appeared first on PHP Architect.

PHP Architect
If Anyone Builds It, Everyone Dies: One Year Closer
IA

If Anyone Builds It, Everyone Dies: One Year Closer

In celebration of still being alive and fighting, we are giving away 1,000 Amazon e-books of “If Anyone Builds It, Everyone Dies”. Feel free to send a copy to yourself, a loved one, or a friend—we need all hands on deck. Today marks exactly one year since If Anyone Builds It, Everyone Dies: Why Superhuman […] The post If Anyone Builds It, Everyone Dies: One Year Closer appeared first on Machine Intelligence Research Institute.

MIRI Blog
GitHub Actions Adds cache-mode to Limit Cache Poisoning Risk
Cybersécurité Programmation

GitHub Actions Adds cache-mode to Limit Cache Poisoning Risk

GitHub has added cache-mode to GitHub Actions, a new setting that limits how workflows and jobs can access the Actions cache. It targets cache poisoning, the technique attackers used to compromise the Ultralytics PyPI package in 2024 and the TanStack npm packages in May 2026. Cache poisoning works because an entry written to the shared Actions cache in one context can be restored and run in another. An attacker who gains write access to a cache key that a trusted workflow later reads can plant…

Socket
Claude Cowork and chat are now one Claude
IA Programmation

Claude Cowork and chat are now one Claude

Claude Cowork and chat are now one Claude In hopefully good news for anyone who, like me, was increasingly confused at Cowork v.s. Claude v.s. Claude Code: Starting today, Claude Cowork and chat are merging into one Claude. Bring a quick question, or hand over a report due at noon, and Claude takes it from there, even after you’ve closed your laptop. [...] This is rolling out to Pro and Max plans first, in the Claude app on web, desktop, and mobile over the coming weeks to existing and new…

Simon Willison's Weblog
Quoting Mustafa Suleyman
IA Programmation

Quoting Mustafa Suleyman

We should not treat models as though they have feelings, preferences, rights, or any entitlement to our welfare. Consciousness is the foundation of our ethical, legal, and political systems. To invite another entity to share any flavor of these rights isn’t justified by the evidence and will make the AI containment and alignment challenge even harder. — Mustafa Suleyman, A warning about ‘model welfare’ Tags: ai-ethics, generative-ai, ai, microsoft, llms

Simon Willison's Weblog