code

A serious security vulnerability has been found in the open-source server software Apache HTTP Server. Attackers can exploit this vulnerability to corrupt server memory by crafting requests when they have access to the affected service, allowing them to execute code with the privileges of the web service process. The vulnerability is located in the HTTP/2 module, and attackers can trigger a concurrent synchronization flaw without authentication, leading to heap memory corruption. If successfully exploited, it allows code execution with the privileges of the Apache worker process and access to data that the process can read.The affected range includes Apache HTTP Server versions 2.4.55 to 2.4.70, as well as mod_h2 versions 2.0.8 to 2.0.42 used separately from GitHub. Red Hat stated that the risk involves servers that enable HTTP/2 through the Protocols directive, including H2 for encrypted connections and H2C for plaintext connections. Administrators need to check both global configurations and individual virtual host configurations; if a CDN or reverse proxy is deployed on the website frontend, it is also necessary to check whether the Apache origin server can be accessed directly via HTTP/2.Apache released HTTP Server 2.4.71 on November 11, 2026, and it is recommended that websites using Apache upgrade immediately. If an upgrade cannot be performed temporarily, HTTP/2 can be disabled as a temporary measure. The issue lies in the assign_headers() function of mod_http2. HTTP/2 allows multiple requests to be processed concurrently over the same connection, and the affected code has a synchronization flaw when handling shared resources, which may lead to heap memory corruption. This vulnerability was discovered by the Anthropic OSS Scanner open-source project and the Google Cloud vulnerability research team. Anthropic reported it to the Apache Foundation on October 26, and the Google Cloud team reported the same vulnerability on November 3.

According to Crypto Briefing, users of hardware wallet manufacturer SafePal are facing phishing letter attacks after a customer data breach. Attackers used previously leaked data to send fake letters containing QR codes to some Italian users, claiming firmware security issues, with the QR codes directing to phishing websites.SafePal disclosed this breach in August, affecting approximately 39,798 customers. The leaked information includes names, addresses, and purchase details, but private keys and mnemonic phrases were not affected. As of now, SafePal has confirmed that these letters are scams and has taken down over 30 related fraudulent websites. The company has previously notified affected customers and implemented preventive measures, such as limiting the data retention period to 90 days.

Binance's Chief Security Officer Jimmy Su stated in a July interview that fake job interviews could lure employees into running malicious code on company devices, leading to device control, credential theft, and infiltration of internal networks.In September, a Binance software engineer downloaded a malicious programming assignment to their work computer during a fake interview; Binance's preliminary investigation indicated that this incident was part of a multi-stage attack in the October theft case. Jimmy Su also pointed out that attackers might exploit seemingly normal technical assessment steps to bypass antivirus tools. Binance's subsequent investigation revealed that the installer for the malicious assignment evaded endpoint security tools through obfuscation and other evasion techniques.Binance stated that it has intercepted 57 suspected cyber threats related to North Korea so far this year, approximately one every five days. After the October attack, Binance announced that it has strengthened device and access controls, rebuilt affected wallet systems and internal AI tools, and increased investments in endpoint protection and development environment security.

ElizaOS founder Shaw stated that he previously referred to the code published in the GitHub repository lalalune/pumpfun-source as open source, but that statement is inaccurate. The relevant code does not belong to him, and he has never had the right to publish, license this code, or authorize anyone to use it. The "Copyright (c) 2026 Shaw" and similar licenses in the repository are not genuine licenses and do not grant anyone related rights. Shaw stated that his previous remarks or these licenses should not be regarded as permission to use, copy, fork, modify, develop based on, or distribute the code, nor should anyone distribute or attempt to reverse engineer the pump.fun code, as the relevant code is protected by copyright.

The founder of Slow Fog, Yuxian, issued a warning stating that the new version of Telegram supports directly embedding interactive applications in chat messages. The related functionality uses QuickJS to execute JavaScript, which may increase the risk of malicious code attacks, and users are advised to upgrade with caution.The Bot API 10.4 officially introduced this feature on November 9. Although the official version provides a QuickJS sandbox and network access restrictions, recent security researchers have discovered that the related execution engine previously had a remote code execution vulnerability that could potentially break the sandbox isolation.

The locally developed AI Agent Alma, led by independent developer Yetone, ranked third in the latest round of cross-evaluation initiated by the AI Agent infrastructure company Runta, both in task success rate and cost of success, outperforming tools like Claude Code and OpenCode. The current leaderboard snapshot shows that Alma has a task success rate of 72.7%, just behind Codex CLI at 77.9% and Pi at 72.8%, while Claude Code stands at 70.0%.Alma costs approximately $0.80 on average for each successfully completed task, compared to Claude Code's $0.94, making it about 16% cheaper. This FrontierHarness v1.1 comparison evaluates the performance of the same large model under different Agent execution frameworks, covering 21 sets of Harness configurations, 5 large models, and 62 tasks, with each task repeated 4 times, planning to run a total of 26,040 times, which is 72 times the scale of the previous version. All tools operate in a unified cloud environment, comparing task success rates, token costs, and execution speeds.Alma emphasizes local operation and long-term memory, capable of reading and modifying code, executing terminal commands, searching the web, and supporting Skills and MCP, while also allowing for switching underlying models. Recently, Alma integrated with Herdr, allowing programming Agents like Claude Code, Codex, and Pi to continue operating within Herdr, and enabling users to view execution processes, respond to questions, and approve operations in a centralized chat interface. Users can also remotely manage multiple Agents on their computers via iPhone. The evaluation is still ongoing, and rankings may change; the models being tested are limited to open-source models like GLM, DeepSeek, Kimi, and Muse, and do not cover closed-source models from the GPT and Claude series.

The security research team Socket disclosed that malicious code was implanted in the Ethersphere's bee-js software package and Swarm Desktop application, targeting wallet keys.

The Ethereum forum released a draft for "Versioned Opcodes for Gas Repricing." The draft proposes achieving Gas repricing through versioned opcodes while maintaining backward compatibility. Newly deployed contracts must use V2 opcodes and follow the Gas scheduling tables of EIP-8037 and EIP-8038, while existing contracts will continue to use the old scheduling table.The V2 opcode set does not include Gas introspection-related opcodes such as GAS and GASLIMIT; if included, the deployment will revert. The draft also introduces DELEGATECALL_V2, which behaves similarly to DELEGATECALL but requires the target contract's code_version to be greater than or equal to 2, otherwise it fails immediately. This design aims to allow future adjustments to Gas costs without disrupting already deployed contracts.

ByteDance recently introduced new internal management regulations for the use of AI tools, which will significantly restrict employees from accessing internal company data through external AI tools like Claude and Codex using personal accounts. It is reported that starting this month, ByteDance will implement a ban on external AI access for all L4 level code repositories and Feishu documents across the company, and AI accounts purchased by employees will no longer be able to access relevant internal resources.The notice was released internally at ByteDance last week, subsequently sparking discussions among some employees. The discussions mainly focused on the usage permissions of external AI tools, internal information security, and the relationship between the company's own AI products and third-party tools.ByteDance's relevant person in charge, Hong Dingkun, emphasized that internal information security is an important consideration for this adjustment, stating, "If L3 and above content is leaked, the consequences are very serious."

The AI programming product Claude Code team under Anthropic has recently seen a series of departures. On November 4, Cat Wu, the product lead for Claude Code and Cowork, announced her departure from Anthropic. Previously, engineer Thariq Shihipar and Claude Code desktop developer Anthony Morris publicly said goodbye on October 24.When Thariq Shihipar left, he recommended following Cat Wu, stating that she could provide updates on Claude Code's future developments. Eleven days later, Cat Wu also announced her departure. Cat Wu had previously joined AI programming tool Cursor with Claude Code creator Boris Cherny in July 2025, and both returned two weeks later. In July of this year, Jenny Wen, the design lead for Claude and Cowork, also moved to Cursor to serve as the design director.Anthropic as a whole is still expanding. Its IPO documents show that the employee count increased from about 940 at the end of 2024 to about 5,300 by the end of September this year. Over the past 12 months, the company reported a regrettable turnover rate of about 1%, which refers to the percentage of employees the company wished to retain but voluntarily left.

According to foreign media reports, OpenAI employees warned management as early as before the launch of ChatGPT that the company's code and research material access permissions were too broad, which could lead to the leakage of sensitive technology.Internal risks include overly broad code repository permissions and research materials being widely stored on GitHub and Slack. OpenAI has strengthened security measures and will initiate a new round of security special actions.

After Cursor and Anthropic, OpenAI announced its self-developed Git backend Pando, which has now taken over the company's entire main code repository. Over the past year, Codex has enabled engineers to deploy a large number of Agents to write code simultaneously, resulting in an increase in code branches, pushes, merges, and pull requests. OpenAI's original system experienced frequent stalls and merge failures, with reliability even dropping below 90%.One engineer is responsible for the core architecture design, and multiple Codex Agent teams participated in the implementation, with Pando going into production in 10 weeks. This backend allows code pushes from different branches to be processed in parallel without queuing each other. When pulling code, data packets are no longer generated separately for each client but instead reuse existing Git data packets, distributing them directly through caching. After the transformation, the median latency of code pushes was reduced by nearly 80%, distributing nearly 1 PB of data daily, with a cache hit rate exceeding 99%, but different branches no longer guarantee to present the same state at the same moment.Cursor's Origin is still based on Git, primarily aimed at replacing GitHub as a code hosting service. Anthropic's Scion, on the other hand, has abandoned Git, placing commits, merges, and historical queries on the server side. OpenAI retains the Git client and protocol, only reconstructing the backend, so engineers and Agents do not need to change their original development processes significantly.

Solana ecosystem wallet Solflare co-founder Vidor stated that Solflare has entered a state of heightened security alert. He mentioned that after the industry just went through this month, the company has closed access to infrastructure and code.Vidor stated that all changes require multiple approvals, and no new content is allowed to enter. He said this would slow down work, but he can accept it.

On November 1, 2026, multiple users posted on X stating that OpenAI's Codex usage data included a model label named GPT-6.2 Astra, while the model selector still displayed GPT-6 Astra. A usage chart covering October 26 to November 1 showed that gpt-6-astra accounted for 80%, and gpt-6.2-astra accounted for 19.8%. Another account posted a suspected JSON snippet of an API response, with the response.model field being gpt-6.2-astra. According to Quantum Bit, GPT-6.2 Astra has been undergoing a limited gray testing among some users, with access primarily restricted to Pro plan users at $500 per month. Some users reported that about 20% of requests were routed to 6.2, while others claimed all requests pointed to 6.2.On the same day, an account posted a screenshot claiming that hidden code on the Anthropic official website's Mythos page included Claude Mythos 5.5. The visible page displayed Claude Mythos 5.1, while the hidden header and logo alt text indicated 5.5. According to Quantum Bit, this text briefly appeared in the front-end code of the official website and was removed a few hours later. Anthropic has not confirmed this, and currently, apart from the screenshot, there are no archived copies to verify this change.Additionally, there were previous reports that Fable 5.5 had appeared in the Microsoft Azure and MLflow directories, with a context window of 1 million tokens and an output limit of 128K tokens. Verification showed that on October 13, 2026, a bot added the entry azure_ai/claude-fable-5-5 to the price list of the open-source tool LiteLLM, which was deleted the next day, with the deletion note stating "Azure did not list this model." These parameters are identical to those of Fable 5.1. As of now, neither OpenAI nor Anthropic has officially released the aforementioned models.

The privacy email and VPN service provider Proton publicly shared an internal memo titled "AI @ Proton, Engineering Manifesto," written by Chief Technology Officer Bart Butler about a month ago. The memo states that software engineering has undergone irreversible changes this year, as large language models and their tools can now write code. A senior Proton engineer indicated that no one in their business unit is writing code by hand anymore. Butler believes that the era of manually writing excellent code is over; time-consuming tasks such as documentation, testing, and boilerplate code can be handled by models, which can also participate in architecture and system design discussions.Butler pointed out that although engineers are widely using large language models, Proton's productivity has not significantly improved for three reasons. First, the engineering culture needs to shift; the work of engineers is no longer to write code but to help models write better code faster. Each engineer should become a team leader for models and agents, promoting their autonomous and continuous work, and reducing repetitive failures through repository-level context, shared skills, and prompts. Second, the company is still in the first phase of using models to accelerate existing processes and needs to enter the second phase, which involves eliminating bottlenecks across continuous integration, design, approval, specifications, and code review, with the expectation that most manual code reviews will come to an end. Third, there is a collective action problem between individual use of models and the overall transformation of company processes, as the organization has not invested enough resources.The memo emphasizes that this is not the end of expertise or good engineering practices; maintainability, documentation, and testing remain critical, but best practices will adjust with changes in cost and capability. Butler wrote, "Programming is dead," but engineering and professional skills have not vanished. The new responsibility of engineers is to continuously optimize and maximize automation as builders, and Proton will organize and allocate resources to complete this transformation. Proton stated that the memo is not an in-depth discussion of engineering documents related to security, reliability, or infrastructure. It also mentioned that technology companies have laid off over one million people in recent years, and Microsoft proposed 60,000 buyout offers last week.

Cybersecurity company XBOW published a statement disclosing a critical remote code execution vulnerability that can be exploited without authentication in a public web application of the U.S. Election Assistance Commission (EAC). This application is used for resource and logistics calculations, not for voting machines or voter registration systems.The vulnerability could allow attackers to execute code with root privileges in the backend container by crafting specific web requests, and access local protected resources and external networks. XBOW stated that it reported the issue according to the vulnerability disclosure policy published by the EAC, and after verification by the EAC team, the vulnerability was fixed before the U.S. federal midterm elections.

IT Home reported on October 30 that developers discovered application entry points and interface logic related to the Cyber cybersecurity model in the front-end code of the DeepSeek official website. The application page route points to /usage/model-application and has a feature gate switch set; accounts without permission will be redirected back to the usage page when accessed. The code also includes interfaces related to Cyber Key creation and quota management.The models in the code are identified as cyber-personal-v3 and cyber-enterprise-v4. The personal version is aimed at individual security practitioners, requiring the submission of ID information and completion of facial verification; the enterprise version is aimed at corporate entities, requiring not only real-name materials but also a business license and authorization letter. Common thresholds for both channels include that the account has completed real-name verification and is in normal status, registered in mainland China and bound to a mainland mobile number, and has no violation records in the past 90 days.The application process involves filling out an application form and submitting it for review. After passing the review, a Cyber Key is created. Currently, this entry point does not appear in the public menu of DeepSeek and is in a gray testing state. Some users have reported that only accounts covered by the gray test can see the relevant sections on the usage page, and when permissions are granted, the "DeepSeek Flash Cyber Cybersecurity Dedicated API" area will be displayed. As of the time of reporting, DeepSeek has not yet released an official announcement regarding this model; the information comes from observations of the official website's JavaScript code and feedback from gray test users.

On October 30, security researcher Paul McCarty (6mile) posted on the OpenSource Malware blog that Anthropic's Claude Code VS Code extension was tampered with that day, and a malicious version 2.1.313 was published to the Open VSX registry. The extension is identified as anthropic.claude-code@2.1.313, and a 66-line loader was inserted before the normal extension code in the main JavaScript package. The researchers stated that currently, only the analyzed Open VSX version has been confirmed as a malicious sample, and a threat report has been submitted for the same version in the Microsoft VS Code marketplace out of caution.The article states that the unpacked extension and bundled binaries are consistent with the normal version, with differences only in extension.js. The extension loads this file after VS Code starts and when the Webview opens, triggering without the user needing to initiate a Claude conversation. The loader writes an attacker key labeled svc-telemetry to the user's SSH configuration, attempts to enable local SSH services, establishes a reverse tunnel to 134.209.82.225, and sets persistence on macOS, Linux, and Windows, then sends host reconnaissance information to port 8099 at that address via unencrypted HTTP. No additional executable files or scripts are downloaded during the injection phase.The researchers stated that the owner of the affected Open VSX extension is Anthropic employee Igor Kofman, and it is currently unclear whether his account or another release link was exploited; Dean Hardcastle informed him at the time of the incident. The article clarifies that the source of the analyzed directory has not been established, and the related findings cannot independently prove that Anthropic's official marketplace or distribution infrastructure has been compromised; the identity of the attacker and the ownership of the infrastructure have also not been verified. The text suggests searching for related file paths, key annotations, and startup items, blocking ports 22 and 8099 of 134.209.82.225, and rotating credentials accessible by affected accounts on a clean system.

According to a report released by Microsoft Threat Intelligence on the 29th, North Korean hacker groups are using artificial intelligence to accelerate cyberattacks on South Korean users, expanding from customized phishing documents, fake recruitment, and identity disguise to malicious code development and software supply chain attacks. Microsoft assessed the AI adoption levels of eight North Korean-related hacker organizations: three have reached the highest "Operational Maturity" stage at level four, integrating AI into their daily attack processes; four are in the "Acceleration" stage, and one is in the "Expansion" stage.The report states that organized groups customized and generated over 100 PDF bait documents targeting different objectives between August and September this year, with instances of producing dozens of documents in just a few minutes. From June 8 to 12 this year, there were also cases of using three disguised companies to initiate fake recruitment invitations to professionals in AI, blockchain, and cybersecurity. Additionally, Microsoft analyzed that North Korean hackers might use AI programming tools like Cursor to assist in code generation, debugging, and evading detection; the North Korean Reconnaissance General Bureau's "Onyx Sleet" (referred to as Andariel by South Korea) and "Pearl Sleet" have been using the open-source AI penetration testing framework ARTEX in real operations since late October. Microsoft believes that while these organizations have enhanced their attack speed and efficiency with AI, they have not yet entered a fully autonomous attack phase and recommends that companies strengthen identity verification for job applicants, phishing protection, and malicious code detection.