30
57
60
42
60
55
60
57
57
60
60
60
45
60
50
60
47
45
35
55
30
55
55
60
30
30
30
30
30
Forgetting Only What Matters: Layer-Selective Unlearning toward Robust LLMs
Researchers introduce FOM-UL, a layer-selective machine unlearning framework that improves forgetting-utility trade-offs and resists knowledge recovery after quantization.
FOM-UL selects transformer layers for unlearning using a forget-to-retain significance score, concentrating parameter updates on layers highly influential for the forget set while leaving most of the model unchanged. It reduces residual memorization versus GA, NPO, KLD, SURE, ReLearn, and LUNAR-based baselines on TOFU, KnowUnDo, and MUSE-style evaluations while preserving retain-set utility. Under 8-bit and 4-bit post-training quantization and adversarial prompts, it maintains stronger suppression of forgotten content, addressing brittleness of diffuse unlearning updates.
36
Re: Vulnerability fixes in util-linux-2.42.3
util-linux 2.42.3 includes vulnerability fixes, with a commit link shared for downstream tracking but no CVEs cited.
An oss-security post links a util-linux commit (286dd3ff41526b582ef48830de239dffbaa61f90) as part of the 2.42.3 vulnerability fix release. No CVE identifiers, flaw descriptions, or exploitation details are provided. The item is a routine open-source fix notification for downstream distributors.
22
Trusting-Trust Attack against an Entire Linux Distribution (via the strip utility)
ArXiv paper shows the trusting-trust compiler backdoor technique can compromise an entire Linux distribution via the strip utility.
The paper (arXiv 2607.24888) demonstrates that Ken Thompson's trusting-trust attack, long viewed as a compiler-specific threat, can backdoor an entire Linux distribution by targeting the strip utility. A compromised tool reproduces its backdoor in subsequent rebuilds of itself, generalizing the attack surface beyond compilers. The finding has supply-chain implications for build reproducibility and distribution trust, though it is a research result with no observed real-world exploitation.
48
EvoSafeHarness: Evolving Model- and Domain-Specific Harnesses for Securing Agents
EvoSafeHarness auto-synthesizes per-model, per-domain safety harnesses, cutting prompt-injection attack success on AgentDojo to 0.0% at 82.8% utility.
EvoSafeHarness is an optimization framework that synthesizes deployable safety harnesses for frozen LLM agents in a target domain, jointly searching natural-language policies and executable code logic guided by model behavior, domain specifications, and adversarial review. On DecodingTrust-Agent it reduces average attack success rate from 45.6% to 10.0% at a 3.3-point utility cost, and on AgentDojo reaches 82.8% utility at 0.0% ASR, twice CaMeL's utility at that operating point. It keeps mean ASR below 20% under adaptive PAIR attacks and transfers unchanged to unseen AgentDyn suites. The analysis finds domain semantics determine required safety relations while model and runtime behavior determine enforcement points.
48
Same Cluster, 33 Points More Utilization: What Changed Was the Order
A Dharma AI blog post on Hugging Face claims GPU cluster utilization rose 33 points after changing job ordering.
A community blog post published on Hugging Face, part of a GPU management series by Dharma AI, discusses improving utilization on the same GPU cluster. According to the title, reordering jobs or tasks was the change that produced roughly 33 additional points of utilization. No article body was available for further detail.
22
Backdoor Xz Utils Linux Open Source
Infosecurity Magazine covers the XZ Utils open-source backdoor, a malicious implant in liblzma that targeted OpenSSH on major Linux distributions.
The article covers the XZ Utils backdoor, a malicious implant introduced into the widely used open-source compression library. The compromised liblzma code manipulated functions used by OpenSSH, nearly reaching stable releases of major Linux distributions before discovery. The incident is a prominent example of software supply chain compromise targeting critical open-source infrastructure.
60
Iran-Linked Hackers Target More US Water Infrastructure in New Jersey and Alabama
Iran-linked hackers hit water utilities in New Jersey and Alabama, bringing confirmed US water-infrastructure attacks to at least 12 states.
Iran-linked hackers attacked the Cape May Sewer Department and Woodbine Water Department in New Jersey and the Childersburg Water, Sewer and Gas system in Alabama on July 27, bringing confirmed US water-sector attacks to at least 12 states since late July. The intrusions targeted internet-exposed industrial control systems, including Rockwell Automation PLCs, with roughly 12 hours of impact in New Jersey and no impact on water quality or service. The FBI confirmed seven affected states by July 30, and CISA has urged utilities to remove PLCs from direct internet exposure. New York announced more than $9 million in grants to strengthen water-sector cybersecurity.
78
30
30
30
45