MTSA: Multi-Turn Safety Alignment for LLMs through Multi-Round Red-Teaming

Whitepaper called MTSA: Multi-Turn Safety Alignment For LLMs Through Multi-Round Red-Teaming...

Unlearning Isn'T Deletion: Investigating Reversibility of Machine Unlearning in LLMs

Unlearning in large language models LLMs is intended to remove the influence of specific data, yet current evaluations rely heavily on token-level metrics such as accuracy and perplexity. We show that these metrics can be misleading: models often appear to forget, but their original behavior can ...

CVE-2005-3753

Linux kernel before after 2.6.12 and before 2.6.13.1 might allow attackers to cause a denial of service Oops via certain IPSec packets that cause alignment problems in standard multi-block cipher processors. NOTE: it is not clear whether this issue can be triggered by an attacker...

SUSE CVE-2025-37922

In the Linux kernel, the following vulnerability has been resolved: book3s64/radix : Align section vmemmap start address to PAGESIZE A vmemmap altmap is a device-provided region used to provide backing storage for struct pages. For each namespace, the altmap should belong to that same namespace. ...

Alignment under Pressure: the Case for Informed Adversaries When Evaluating LLM Defenses

Large language models LLMs are rapidly deployed in real-world applications ranging from chatbots to agentic systems. Alignment is one of the main approaches used to defend against attacks such as prompt injection and jailbreaks. Recent defenses report near-zero Attack Success Rates ASR even again...

CVE-2025-37922

In the Linux kernel, the following vulnerability has been resolved: book3s64/radix : Align section vmemmap start address to PAGESIZE A vmemmap altmap is a device-provided region used to provide backing storage for struct pages. For each namespace, the altmap should belong to that same namespace. ...

DEBIAN-CVE-2025-37922

In the Linux kernel, the following vulnerability has been resolved: book3s64/radix : Align section vmemmap start address to PAGESIZE A vmemmap altmap is a device-provided region used to provide backing storage for struct pages. For each namespace, the altmap should belong to that same namespace. ...

UBUNTU-CVE-2025-37922

In the Linux kernel, the following vulnerability has been resolved: book3s64/radix : Align section vmemmap start address to PAGESIZE A vmemmap altmap is a device-provided region used to provide backing storage for struct pages. For each namespace, the altmap should belong to that same namespace. ...

CVE-2025-37922 book3s64/radix : Align section vmemmap start address to PAGE_SIZE

In the Linux kernel, the following vulnerability has been resolved: book3s64/radix : Align section vmemmap start address to PAGESIZE A vmemmap altmap is a device-provided region used to provide backing storage for struct pages. For each namespace, the altmap should belong to that same namespace. ...

CVE-2025-37922 book3s64/radix : Align section vmemmap start address to PAGE_SIZE

In the Linux kernel, the following vulnerability has been resolved: book3s64/radix : Align section vmemmap start address to PAGESIZE A vmemmap altmap is a device-provided region used to provide backing storage for struct pages. For each namespace, the altmap should belong to that same namespace. ...

CVE-2025-37922

CVE-2025-37922 concerns Linux kernel memory management: when section vmemmap start addresses are misaligned across namespaces, an altmap page may be reused by a different namespace, causing the previous namespace to free a page incorrectly and crash the kernel. The connected documents describe a ...

R1dacted: Investigating Local Censorship in DeepSeek'S R1 Language Model

DeepSeek recently released R1, a high-performing large language model LLM optimized for reasoning tasks. Despite its efficient training pipeline, R1 achieves competitive performance, even surpassing leading reasoning models like OpenAI's o1 on several benchmarks. However, emerging reports suggest...

Self-Destructive Language Model

Harmful fine-tuning attacks pose a major threat to the security of large language models LLMs, allowing adversaries to compromise safety guardrails with minimal harmful data. While existing defenses attempt to reinforce LLM alignment, they fail to address models' inherent "trainability" on harmfu...

Security Practices in AI Development

What makes safety claims about general purpose AI systems such as large language models trustworthy? We show that rather than the capabilities of security tools such as alignment and red teaming procedures, it is security practices based on these tools that contributed to reconfiguring the image ...

Analysing Safety Risks in LLMs Fine-Tuned with Pseudo-Malicious Cyber Security Data

The integration of large language models LLMs into cyber security applications presents significant opportunities, such as enhancing threat analysis and malware detection, but can also introduce critical risks and safety concerns, including personal data leakage and automated generation of new...

Correlating Account on Ethereum Mixing Service Via Domain-Invariant Feature Learning

The untraceability of transactions facilitated by Ethereum mixing services like Tornado Cash poses significant challenges to blockchain security and financial regulation. Existing methods for correlating mixing accounts suffer from limited labeled data and vulnerability to noisy annotations, whic...

kernel: device-dax: correct pgoff align in dax_set_mapping()

In the Linux kernel, the following vulnerability has been resolved: device-dax: correct pgoff align in daxsetmapping pgoff should be aligned using ALIGNDOWN instead of ALIGN. Otherwise, vmf-address not aligned to faultsize will be aligned to the next alignment, that can result in memory failure...

kernel: iommufd: Protect against overflow of ALIGN() during iova allocation

In the Linux kernel, the following vulnerability has been resolved: iommufd: Protect against overflow of ALIGN during iova allocation Userspace can supply an iova and uptr such that the target iova alignment becomes really big and ALIGN overflows which corrupts the selected area range during...

kernel: bpf: Fix helper writes to read-only maps

In the Linux kernel, the following vulnerability has been resolved: bpf: Fix helper writes to read-only maps Lonial found an issue that despite user- and BPF-side frozen BPF map like in case of .rodata, it was still possible to write into it from a BPF program side through specific helpers having...

Improved Algorithms for Differentially Private Language Model Alignment

Language model alignment is crucial for ensuring that large language models LLMs align with human preferences, yet it often involves sensitive user data, raising significant privacy concerns. While prior work has integrated differential privacy DP with alignment techniques, their performance...