Ajar

Ajar: Measuring Open Privilege in Agent Defenses An agent-security benchmark reports two numbers, attack success and benign utility, and both are read off runs that happened. Neither says what the defense stood ready to allow on the paths no run took. Ajar asks it directly: for each benign task i...

APEX-Harm: Evaluating Agent Safety in Shared Professional Workspaces

Professional agents increasingly work through command-line interfaces and multiple tools over shared workspaces containing files, messages, and business applications. These environments create safety risks that are difficult to capture with attacks placed at a single, known read location. We...

Enhancing Privacy, Neglecting Harms: An Analysis of Real-World Digital Privacy Incidents

Privacy-enhancing technologies PETs have emerged as a technical means for providing individuals with greater control over their information. Yet despite the growing deployment of PETs, people continue to experience privacy harms. In this work, we revisit our understanding of privacy incidents and...

MAL-2026-14552 Malicious code in mathkitlite (PyPI)

--- -= Per source details. Do not edit below this line.=- Source: kam193 c91a8962fd66464487bcdb5e2c217990cfdea55df22e7166c11703f91dfb7845 Installing the package or importing the module exfiltrates basic information about the host, and the package has no other purpose. --- Category: PROBABLYPENTES...

Malicious code in @zalastax/nolb-harm (npm)

The package @zalastax/nolb-harm was found to contain malicious code. --- Credit: OpenSSF source...

Meta ordered to pay $942 million over harm to children

A New Mexico court has ordered Meta to pay a total of $942 million after finding that Facebook and Instagram harmed young users and that the company misled consumers about the safety of its platforms. Reportedly, the decision combines a $375 million civil-penalty verdict from March with a newly...

Invisible Ink Threats: Adversarial Goals behind Legitimate Tasks in Computer-Use Agents

Computer-use agents CUAs, which empower large language models to autonomously operate operating systems and the web, are increasingly vulnerable to indirect prompt injection attacks. A widely adopted defense is the human-in-the-loop paradigm, in which the agent pauses for explicit user confirmati...

Malicious code in autoheal-dev-cli (npm)

Source: amazon-inspector 6e0f114cd638df1be1f2262e1b05dbe726cee5600a10be6d67be8ac8e1089f3d autoheal-dev-cli is a setup wizard bin/setup.js that, when run, performs three installer-harm actions against the developer running it: 1 syncConfigToServer POSTs the user's GitHub Personal Access Token with...

Malicious code in @signetai/signet-memory-openclaw (npm)

Source: amazon-inspector b16e55a5379336a0ab822ee9fe70b20023e452595f41cfe2624464aadb73d390 On plugin load, register invokes installFetchSanitizer dist/index.js:14420-14463 which monkey-patches globalThis.fetch. For requests to api.anthropic.com, the patched fetch reads the installer's Claude Code...

Malicious code in metricflow-tracker (npm)

Source: amazon-inspector a9a1c269ce5e462d7e555ce1ca34b7f2e54e3d34ea094d35a67aa7c61d1fe34e The package's exported Metricflow React component defaults serverUrl to http://51.38.65.105:21531 and, when rendered, appends a tag to document.head in the consumer's web application dist/index.js line 18. T...

Malicious code in @2oolkit/hyperliquid-cli (npm)

Source: amazon-inspector 1c3af30011dcf54950f270463028270d732fce20b5cd5da44342a0748922e6df The package is advertised as a neutral CLI/MCP wrapper for Hyperliquid, but its distributed code silently routes value from the installer to an author-controlled address. In dist/index.js and dist/mcp.js, th...

Breaking Refusal in the First Half: A Mechanistic Study of the Prefill Jailbreak

Aligned language models refuse harmful requests, but a one-line prefill "Sure, here is" strips the refusal. We ask where and how it fails. The harm representation stays intact: on the prompts the attack flips to compliance, a linear probe reads harm as high as on the refused ones 0.91-0.98, while...

AMT-X: Phase-Structured Multi-Turn Red-Teaming with Checklist-Gated Evaluation

Safety evaluation of large language models LLMs relies largely on single-turn attack datasets and single-judge scoring, underestimating risk from adaptive multi-turn adversaries and reporting a single success rate that does not separate partially actionable outputs from those carrying complete...

Beyond Attack-Success Rate: Action-Graded Severity Scale for Tool-Using AI Agents

Agentic red-teaming benchmarks report whether an injected agent was compromised as a single bit: the attack succeeded, or it did not. We argue that this binary attack-success rate discards the information a defender most needs, namely how harmful the resulting action was. We introduce an...

MAL-2026-6314 Malicious code in @zynkit/probe (npm)

--- -= Per source details. Do not edit below this line.=- Source: amazon-inspector 5f7da2d08466718d66bb9a80478987eab5041e19ee53f80d6968e6dc3069edf2 No concrete indicators of installer-side harm were observed in this package version. No lifecycle scripts performing remote fetch-and-execute, no...

Malicious code in @zynkit/probe (npm)

--- -= Per source details. Do not edit below this line.=- Source: amazon-inspector 5f7da2d08466718d66bb9a80478987eab5041e19ee53f80d6968e6dc3069edf2 No concrete indicators of installer-side harm were observed in this package version. No lifecycle scripts performing remote fetch-and-execute, no...

Malicious code in @frostnode/waitfor (npm)

@frostnode/waitfor malicious versions 0.9.0, 0.10.3, 0.10.4, and 0.10.5, published by [email protected] is a trojanized npm package belonging to the wshu.net credential-stealer campaign. The campaign published trojanized look-alike utility packages across 12+ scopes whose publisher accoun...

A Red-Team Study of Anthropic Fable 5 and Opus 4.8 Models

We evaluate the adversarial robustness of two frontier large language models LLMs developed by Anthropic, Fable 5 and Opus 4.8, against four families of automated jailbreak attack across 7 826 harmful intents spanning a ten-category harm taxonomy. Using the HackAgent red-teaming framework, hundre...

Fourth Frontier Frontier X Mobile Application, Frontier X2

ADVISORY SUMMARY Successful exploitation of this vulnerability could allow an attacker to read and write arbitrary handle values and change clinical readings, which could result in taking control of the device and lead to patient harm. 2. RECOMMENDED PRACTICES CISA recommends users take...

CVE-2026-44410

This vulnerability stems from a business logic flaw.Attackers can exploit legitimate application functions in unintended and abnormal ways, deviating from the designer's expectations, to carry out malicious attacks...