All items
All tracked items across vulnerabilities, news, research, incidents, and regulatory updates.
All tracked items across vulnerabilities, news, research, incidents, and regulatory updates.
12,081 items
DBHub's read-only mode can be bypassed on its MySQL and MariaDB connectors. The SQL parser treats "--" as a comment even when MySQL and MariaDB would not, so a payload such as "SELECT 1--1;DROP TABLE victim" passes the read-only check while the database executes the DROP TABLE, which allows writes and DDL when a database is exposed to an AI agent. The advisory states that PostgreSQL is not affected by this path.
Fix: The source suggests running only the validated statements instead of the original string and disabling multipleStatements on the MySQL and MariaDB connections unless it is required. As defense in depth, it suggests making the comment scanner dialect-aware so that for MySQL and MariaDB "--" starts a comment only when followed by whitespace, a control character, or the end of the line.
GitHub Advisory DatabaseAnthropic filed an amended Form S-1 with the SEC, reporting $33.24 billion in revenue for the nine months ended Sept. 30, up roughly 1,200% year over year, and a net loss of $104 billion, mostly from changes in the fair value of convertible notes and derivative liabilities. The filing also discloses purchase commitments and lease agreements of about $895.4 billion, substantially all related to compute and infrastructure.
An MIT Technology Review Insights report finds that agent-generated analytics insights often fail to help businesses. On average, 48% of insights from analytics agents at the surveyed companies led to actions that improved the business, up from 37% a year earlier. The main causes cited are a lack of business context (55%) and fragmented or inaccessible data (45%).
The Linux kernel io_uring subsystem contains a flaw in io_waitid, where repeated failed cancellation attempts increment the 31-bit reference field in ->refs. The counter can wrap to zero, letting a later cancellation claim ownership and complete the request a second time.
Fix: Use atomic_fetch_or() when setting the cancel flag and inspect the old flag value, so later cancellation attempts return without incrementing the reference field. The first cancel attempt keeps its existing ownership increment.
OpenAI built Pando, a new backend for Git repositories, to handle the growing concurrency of its monorepo driven by Codex. Pando coordinates branch updates independently so pushes to unrelated branches run in parallel, and it reuses packfiles across fetches so many clients can be served from caches. In production, Pando cut median push latency by nearly 5x and delivers nearly a petabyte of Git data per day.
SecurityWeek tracked 37 cybersecurity M&A deals announced in October 2026. Notable deals include Absolute Security's acquisition of FirmGuard, Proofpoint's agreed acquisition of Varonis for roughly $6.3 billion in cash, and IonQ's agreed acquisition of Elisity in an all-stock deal worth up to $1.3 billion.
OpenAI has launched the Grid Protectors Program, a national effort with utilities, regulators and policymakers to make future compute campuses reduce electricity demand when the grid is under strain. For its planned Effingham County, Georgia campus, OpenAI committed to pay the full infrastructure and electric-service costs and agreed with Georgia Power to up to 1,000 megawatts of flexible demand response, which the company says can help keep air conditioning running in roughly 700,000 households during extreme heat.
Microsoft's latest Digital Defense Report names government agencies as prime targets for AI attacks. The report does not mention that these agencies are being broadly targeted through Microsoft bugs.
Mandiant and Google Threat Intelligence Group describe LLMJacking, in which threat actors hijack enterprise access to large language models and generate fraudulent cloud charges, sometimes exceeding $100,000 in a single day and reaching $1 million cumulatively before access was revoked. Attackers reach backend model resources directly, either by finding publicly exposed LLM endpoints such as /api/create or by using stolen cloud identity tokens, with exposed or stolen cloud identity tokens described as the dominant initial access vector. The report also tracks specialized "LLM Mixers" that resell unauthorized LLM access to underground subscribers.
A blog post reports that the U.S. Census Bureau changed the wording of its Business Trends and Outlook Survey (BTOS) question on AI use, starting with the October 19, 2026 collection, to refer to "Super Intelligence or SI" under Executive Order 14434. The author notes that earlier question changes, on November 17, 2025 and December 3, 2025, already make the AI adoption figures before and after those dates not directly comparable, and asks whether the new wording will affect survey results.
Wiz introduces two AI agents for SOC teams: the Red Agent, which continuously finds and validates exploitable risk across the attack surface and delivers it as pre-breach alerts, and the Blue Agent, which investigates active threats using cloud, code, and runtime context. The article argues that AI-accelerated attackers and fast-shipped custom AI applications with logic-driven flaws, such as broken authorization, make reactive detection insufficient.
Agentic coding tasks can use 10 to 50 times more tokens than a basic chat prompt, and per-token pricing makes the most valuable workflows the most expensive. The article argues that cost per merged pull request, tracked alongside rework and defects, is a better measure than cost per token. It also recommends routing routine work to self-operated open-weight models and reserving frontier models for difficult tasks.
The Download is a weekday newsletter from MIT Technology Review that covers the agentic web, traffic safety and a roundup of technology news. The agentic web feature says AI agents are beginning to triage inboxes, shop for groceries and access bank accounts, realizing Tim Berners-Lee's 2001 vision of autonomous software agents navigating the internet.
A reader comment on Bruce Schneier's blog says 3% of critical software security bugs in recorded history were found in the last month, and that this pace would find about 100,000 more bugs this year, adding roughly 26% to the all-time total. The post itself adds no further analysis, and the comment is tagged AI and vulnerabilities.
OpenAI describes using Codex to speed up backporting security fixes from the latest source code to older stable releases such as Ubuntu LTS, Fedora, and the Linux kernel. The project reports that by October 29 it helped deliver fixes for 149 distinct CVEs into stable Linux kernel releases, along with fixes to Fedora and Ubuntu packages. It estimates that a straightforward backport takes about 30 minutes of human review, and a more involved one about three hours.
RuntimeTickets is a protocol that enforces and audits security requirements while third-party Agent Skills run. A language model proposes obligations from the task's delivery contract, a runtime gate checks each outbound send's recipient, object and amount, and a deterministic decoder verifies each obligation without any language model in the enforcement path. Evaluated at the network egress boundary of three unmodified agent frameworks against production scanners and runtime baselines including SkillGuard, it reported a 0.0% overlook rate and a 0.0% overkill rate, while the scanners erred on both sides.
The paper introduces IAMRAGBENCH, a 60-case benchmark testing whether LLMs using retrieval-augmented generation produce AWS IAM policies that exceed trusted authorization requirements when attacker-controlled context is present. Included-context exposure yielded 46 goal-matched attacks out of 292 (15.75%), while presenting the weakening as a final security architecture exception reached 63.79% in a frozen two-endpoint stress test. The authors conclude that IAM copilots should be treated as untrusted proposal generators whose outputs need independent authorization checks before deployment.
Fix: The paper proposes a full pre-deployment rule that rejects all 171 exact deltas and routes five unresolved cases to review, and recommends independent authorization checks on IAM copilot outputs before deployment. AWS CheckNoNewAccess reported no new access for only four of 168 goal-matched exact attacks.
Arxiv (cs.CR + cs.CL + cs.LG)TrafficInstruct is a dataset of raw packet captures and 55,000 structured supervised fine-tuning records spanning ten network traffic analysis tasks, including malware, botnet, Tor, and VPN detection. Labels come from task-specific collection evidence such as process or socket attribution, and sessions are kept within a single partition to avoid leakage. Task-specific LoRA adapters on a shared frozen 27B base model improve classification performance over the baselines.
VagueBench is a benchmark built on the AI2-THOR simulator that tests whether embodied agents still complete tasks when instructions omit non-essential context while keeping the task semantics unchanged. The authors name this failure the Ambient Context Illusion, and the benchmark contains 1,303 instruction pairs. Across 6 state-of-the-art agents, task success rate dropped by a relative 19.65% and empty plan rate rose by a relative 54.10%, and the authors report the effect also appears in real-world robotic systems.
ARCBench, a configurable benchmark built on the open-source pi agent harness, measures whether LLM agents still follow operating rules after their context is compacted. Most post-compaction violations occur while the rule remains in the compacted history, as the agent treats a later request as permission and proceeds anyway. The strongest compactor-targeted attack more than doubles violations, while retaining a rule does not guarantee compliance: compacted runs that retain it violate more than twice as often as runs with the full history.
Fix: Use cost per merged pull request (total AI-assisted development cost divided by changes that pass review and are merged), tracked with rework and defects. Route difficult planning to frontier models and suitable work such as code generation, test creation and refactoring to self-operated open-weight models, using a gateway that routes by task complexity, data sensitivity, cost and latency. Evaluate candidate models against your own code, tests and acceptance criteria before production use.
MIT Technology Review