OpenAI Google and Microsoft lead AI news

Calif says it used OpenAI’s GPT-6 Cyber model to discover Java4Shell, a memory-corruption vulnerability in OpenJDK’s Base64 decoder that it says may have existed for more than a decade. Researcher Quang Luong reportedly found the flaw in about an hour and built an exploit in two more, using less than $200 in AI credits. Certain x86-64 systems are affected; the vulnerable code path does not affect ARM64. A crafted Base64 input can crash a Java virtual machine and, under some conditions, enable remote code execution. Calif reports demonstrating pre-authentication execution on Atlassian Jira, Jenkins and Keycloak, and post-authentication execution on Apache Kafka Broker and Spinnaker. It says not every Java application is remotely exploitable and researchers do not know whether attackers are exploiting the flaw.

Java is used by about 90% of Fortune 500 companies and across sectors including finance, healthcare, aviation, telecommunications and government. Calif coordinated with Five Eyes partners, vendors including Google and Microsoft, and cloud providers; European authorities are also coordinating responses. Oracle released a critical patch update, while CERT-EU identified affected versions and fixed builds across seven OpenJDK distributions. CISA guidance advises organizations to inventory bundled Java runtimes and prioritize internet-facing services. Where patching is not immediately possible, researchers say a workaround is to disable the affected Base64 encoder. Calif has also reported finding serious PHP vulnerabilities; one article says its earlier PHP finding affected about 30 million websites. The articles differ in the level of technical detail they provide, and Oracle had not publicly commented specifically on Java4Shell in some accounts.

OpenAI’s demos show PR scans within a pull-request workflow and the Codex Security Plugin, including how users start it and review results. Separately, Groundcover has launched Agent Investigations, a shared workspace for engineers and AI agents to investigate production incidents using logs, metrics, traces and Kubernetes data. Its Investigator agent tests hypotheses and a Moderator agent checks evidence; investigations can begin automatically from monitor alerts or through Groundcover and Slack. Anthropic’s Claude Code and OpenAI’s Codex can also join through Groundcover’s Model Context Protocol server. The service is available to Groundcover customers with AI enabled on their backends, and the company plans to demonstrate it at AWS re:Invent, November 30–December 4.

Strong demand for Anthropic’s IPO lifted U.S. stock futures and renewed interest in AI shares. One report says the offering sold 980 million shares at $83 each, raising about $81.3 billion; another gives a target valuation of roughly $1.4 trillion. December S&P 500 E-Mini futures rose 1.21% and Nasdaq 100 futures rose 2.20% in one morning report, while another put Nasdaq futures up 1.9%. OpenAI chairman Bret Taylor welcomed a public listing as a way for more investors to participate in AI’s growth, while warning against excessive risk. A Seeking Alpha analyst’s bullish case cites business growth and improving computing efficiency, but flags $895.4 billion in commitments, including about $447.5 billion due before 2031, alongside risks from competition, cheaper models and slower adoption. The articles also caution that infrastructure costs and financing needs remain concerns.

Astera Labs, which makes connectivity products for AI systems, reported quarterly revenue of $584.9 million, up 154% year over year and 49% from the prior quarter, with growth in its Scorpio family after X-Series switches ramped up. Its Amazon-related ecosystem could offer indirect exposure to computing demand associated with Anthropic, but the article reports no direct Anthropic sales. Customer concentration remains a risk, and Amazon-related warrant conditions do not guarantee orders.

Companies are also confronting the cost of AI use as vendors move from flat per-seat pricing toward token-based charges. Wiv CEO Roni Karp says some businesses pay $800–$1,000 monthly per developer for AI tools. Huron Consulting Group’s Luis Solano says token prices have declined since a mid-July peak, but total spending keeps rising as firms use AI for more tasks. Google and Microsoft have reported sharp growth in AI use, while energy shortages could constrain computing capacity. Experts say companies are still working out how to control costs and select suitable models; Elon Musk has warned that AI agents could require 1,000 times more bandwidth.

Delaware State Police added AI translation to body cameras, allowing officers to speak in more than 50 languages and hear replies translated into English. The state included the service in a renewed contract, but officials did not disclose its cost; smaller departments in Cheswold, Fenwick Island and Georgetown cite expense as a possible barrier. Certified translators can review transcripts when needed, and each department decides whether to purchase the software. In a separate product test, a writer who used Apple’s Siri Recap on an Apple Watch Ultra 4 for two weeks found it useful for recalling meetings and connecting interview recaps with Voice Memos, but sometimes misleading when it treated video audio as personal events. The writer preferred its privacy approach to Amazon Bee Frontier, whose stored transcripts and summaries prompted deletion of the writer’s data.

Key Takeaways

  • Calif used OpenAI’s GPT-6 Cyber to find Java4Shell in OpenJDK’s Base64 decoder; one researcher reportedly found it in about an hour using less than $200 in AI credits.
  • The flaw affects certain x86-64 systems, not the affected code path on ARM64, and can enable remote code execution under some conditions; Calif demonstrated exploits against Jira, Jenkins, Keycloak, Kafka Broker and Spinnaker.
  • Oracle released a critical patch update. CERT-EU identified affected versions and fixed builds, while CISA advises inventorying bundled Java runtimes and prioritizing internet-facing systems; researchers also describe disabling the affected encoder as a workaround.
  • Calif coordinated disclosure with Five Eyes partners, Google, Microsoft, cloud providers and European cybersecurity authorities; it has also reported serious PHP vulnerabilities.
  • Anthropic’s IPO reportedly raised about $81.3 billion from 980 million shares priced at $83; reports cite a target valuation near $1.4 trillion and renewed investor enthusiasm for AI stocks.
  • OpenAI chairman Bret Taylor welcomed Anthropic’s planned listing but warned investors against excessive risk. A bullish analyst also highlights Anthropic’s $895.4 billion in commitments and risks from competition and slower adoption.
  • Astera Labs reported $584.9 million in quarterly revenue, up 154% year over year; its Amazon ecosystem exposure is indirect, with customer concentration risks and no guaranteed orders.
  • AI software costs are increasingly tied to token use; some firms reportedly pay $800–$1,000 per developer each month. Google and Microsoft report rising AI use, while energy and bandwidth needs remain concerns.
  • Delaware State Police added live two-way AI translation in more than 50 languages to body cameras; smaller departments cite cost as a barrier, and the state has not disclosed its contract cost.
  • Groundcover launched a shared AI-and-engineer incident investigation workspace that can use Anthropic Claude Code and OpenAI Codex. Apple’s Siri Recap test found some meeting-recall benefits but occasional misinterpretation of video audio.

AI uncovers serious Java security flaw used across major industries

Cybersecurity startup Calif used OpenAI’s GPT-6 Cyber model to find a serious flaw in Java, which is used by about 90% of Fortune 500 companies and by banks, hospitals, aviation firms and government agencies. The flaw could let an attacker gain remote access to sensitive systems. Calif worked with the Five Eyes intelligence-sharing alliance and private vendors, including Google and Microsoft, to coordinate a response. Oracle released a software patch on Tuesday, but it did not comment on the flaw. Calif also recently found serious vulnerabilities in PHP, another widely used programming language.

AI finds Java flaw in under an hour for less than $200

Calif researcher Quang Luong found Java4Shell, a memory-corruption flaw in OpenJDK’s Base64 decoder, in about an hour using $200 in computing credits. Tracked as CVE-2026-114938, the flaw affects certain x86-64 systems and Java branches 17, 21, 25 and 27; ARM64 systems are not affected by this code path. A specially crafted Base64 input can crash a Java virtual machine and may allow code execution under certain conditions, though not every Java application is remotely exploitable. CERT-EU identified affected versions and fixed builds, and listed seven major OpenJDK distributions that carry the exposure. CISA guidance calls for an inventory of bundled Java runtimes and prioritizing updates for internet-facing services.

AI helps uncover a Java flaw that may have gone unnoticed for years

Startup Calif used artificial intelligence to find Java4Shell, a serious flaw in Java that the company said had existed for more than a decade. The flaw could let an attacker remotely run unwanted commands through Java’s memory system. Java is used in critical fields including banking, national security, aviation and telecommunications. Oracle released a critical patch update on Tuesday, but had not publicly acknowledged the specific flaw, according to the article. Computer scientist Professor Toby Walsh said AI is finding software weaknesses faster, including flaws that may have been too difficult for people to discover alone.

AI uncovers critical Java flaw as companies prepare to patch systems

A technology startup used artificial intelligence to find a critical security flaw in Java, one of the world’s most widely used programming languages. The company had also found a major flaw in PHP, showing how AI can help identify weaknesses in software. The article says the Java issue could affect widely used business systems, but provides no specific details about the flaw or affected versions. It notes that AI tools can analyze large amounts of code faster than traditional manual reviews and testing. Developers and enterprise security teams were awaiting information about affected versions and patches.

AI finds Java4Shell flaw in software used by banks and governments

Cybersecurity firm Calif used OpenAI’s GPT-6 Cyber model to uncover Java4Shell, a serious flaw that the company said had existed for more than a decade. The vulnerability could let an attacker remotely run unwanted commands through Java’s memory system. Java is used by major companies, banks, hospitals and governments, and experts warned the flaw could put financial systems and data at risk. Oracle released a critical patch update on Tuesday, but the article says it was unclear whether any platforms or companies had been affected. Calif’s discovery followed its earlier finding of a major PHP flaw, which experts said affected about 30 million websites.

AI uncovers Java flaw that could give hackers administrator access

Security firm Calif used AI to find Java4Shell, a memory corruption flaw in OpenJDK’s Base64 decoder. The bug affects a faster decoding process that uses x86-64 CPU instructions. Attackers could send malformed code to overwrite memory, gain administrator-level access, run malware and expose sensitive data. The flaw could put Java platforms and services at risk.

AI finds Java4Shell flaw that can enable remote code execution

Researchers at US security firm Calif used OpenAI’s GPT-6 Cyber model to discover Java4Shell, a flaw in the Base64 decoder in OpenJDK. Malformed inputs can overwrite nearby memory, potentially crashing apps, exposing secrets or allowing remote code execution. Calif said it achieved pre-authentication remote code execution on Atlassian Jira, Jenkins and Keycloak, and post-authentication execution on Apache Kafka Broker and Spinnaker. The flaw requires an x86-64 processor with AVX-512 VBMI support, and ARM64 CPUs are not affected. Organizations unable to install a patch can use workarounds that disable the affected Base64 encoder; researchers said they do not know whether attackers are exploiting the bug. OpenAI’s model launched on November 5 to a select group of approved security clients.

Java4Shell flaw puts widely used business systems at risk

Security startup Calif used AI to find Java4Shell, a memory corruption flaw in OpenJDK’s Base64 decoder. Attackers could exploit it to take over a vulnerable server over the internet without a password or login. Java is used by 90% of Fortune 500 companies and supports services in finance, cloud computing, aviation, telecommunications, healthcare and more. Before announcing the discovery, Calif coordinated with Five Eyes intelligence partners and major cloud providers so defenders could prepare patches. European cybersecurity authorities are also coordinating responses as Oracle and other maintainers roll out fixes.

AI helps uncover Java flaw with a low-cost exploit

Security firm Calif said it used OpenAI’s GPT-6 Cyber model to find Java4Shell, a flaw in Java’s Base64 decoder. Researchers found that decoding 64 “A” characters followed by an equal sign could corrupt memory and allow access to the system, sensitive data or remote code execution. Calif said one researcher spent an hour finding the bug and two hours building an exploit, using less than $200 in AI tokens. Java is used by about 90% of Fortune 500 companies, and Calif said the flaw could be used to break into Keycloak, an open-source login system used by organizations including NASA and UnitedHealthcare. Calif researchers describe the human and AI partnership as a “centaur hacker” and warn that future models may work without a human handler.

Java security flaw can be triggered by a string of As

Researchers at cybersecurity startup Calif found a Java flaw using AI. The bug can be triggered by input beginning with a long string of “A” characters and could let hackers access systems protected by the Keycloak login system, which is used by organizations including CERN and NASA. Java supports platforms used by banks, hospitals, airlines and government agencies, and about 90% of Fortune 500 companies use it. Calif said the flaw is not easy to exploit, giving defenders some time to respond. The article also notes that recent outages have raised concerns about aging digital infrastructure, though officials have not linked those outages to AI.

Watch a walkthrough of PR scans in a pull request

OpenAI’s product demo shows PR scans in action. The presenter reviews scan results as part of a pull request workflow.

See how the Codex Security Plugin works

OpenAI’s demo walks through the Codex Security Plugin workflow. It shows how to get started and review the results displayed in the recording.

Anthropic IPO lifts U.S. stock futures and AI optimism

U.S. stock futures rose as strong demand for Anthropic’s initial public offering renewed investor enthusiasm for AI stocks. December S&P 500 E-Mini futures (ESZ26) gained 1.21%, while December Nasdaq 100 E-Mini futures (NQZ26) rose 2.20% that morning.

Anthropic IPO boosts Wall Street futures and AI stocks

U.S. stock futures pointed to a higher opening on Thursday, with technology shares expected to lead gains. Nasdaq 100 futures rose 1.9% after investors responded strongly to the IPO of Anthropic, the developer of the Claude AI platform. The offering reportedly sold 980 million shares at $83 each and raised about $81.3 billion, exceeding early pricing expectations. The IPO renewed interest in AI investments, though concerns remain about the industry’s infrastructure costs and financing needs.

Astera Labs expands its role in AI connectivity

Astera Labs makes products that move data between processors, memory, and other parts of AI systems. Its role in Amazon’s infrastructure ecosystem may give it indirect exposure to growing demand for computing linked to Anthropic, though the article does not report direct sales to Anthropic. Quarterly revenue reached $584.9 million, up 154% from a year earlier and 49% from the previous quarter, as its Scorpio product family grew after the X-Series switches ramped up. The article also notes customer concentration risks and says Amazon-related warrant conditions do not guarantee orders. A successful Anthropic offering could support a major customer’s finances, but would not resolve questions about AI spending and returns.

AI token pricing is driving corporate software costs higher

Major AI vendors have moved from flat, per-seat fees to charging companies based on token use, and some businesses now pay $800 to $1,000 a month per developer for AI tools, says Wiv CEO Roni Karp. Huron Consulting Group’s Luis Solano says token prices have fallen since a mid-July peak, but total spending continues to rise as companies use AI for more tasks. Elon Musk warned that AI agents could need 1,000 times more bandwidth, while Google and Microsoft reported sharp growth in AI use. Energy shortages could limit computing capacity, and experts say companies are still learning how to manage costs and choose the right models for each task.

OpenAI chairman says Anthropic IPO shows AI market maturity but warns of risks

OpenAI chairman Bret Taylor welcomed Anthropic’s plan to go public, saying it would let more investors share in AI’s growth. Taylor is also the co-founder and CEO of enterprise AI company Sierra. Anthropic, which makes the Claude large language model, is set to debut on the Nasdaq on Thursday in a listing that could rank among the largest technology offerings in history. The company is targeting a valuation of about US$1.4 trillion, while Taylor cautioned investors to separate sound decisions from moves that take excessive risks.

Analyst backs Anthropic IPO despite huge commitments and valuation risks

A Seeking Alpha analyst rates Anthropic, PBC a Buy, arguing that strong business growth and improving computing efficiency could support its planned IPO valuation. At the proposed high-end price, the company would be valued at $1.3 trillion, or 19 times its annualized revenue; 100 percent growth could bring that multiple below 10 times sales. The analyst says Anthropic faces $895.4 billion in commitments, including about $447.5 billion due before 2031. Cheaper models, strong competitors and slower business adoption could hurt growth, cash availability and valuation, while the company’s switching costs and lasting advantages remain uncertain.

Delaware troopers add AI translation to cameras as smaller departments cite cost

The Delaware State Police has added AI software to officers’ body cameras for live, two-way translation in more than 50 languages. Officers can speak into the camera and play their words in another language, while the software can translate a person’s reply into English. The state paid for the service as part of a renewed contract, but officials did not provide the cost; the article says a Freedom of Information Act request is needed to obtain it. Several smaller departments, including those in Cheswold, Fenwick Island and Georgetown, said high costs could prevent them from adopting the tool. Delaware State Police says certified translators will review transcripts as needed, while each department decides whether to buy the software.

Two weeks with Siri Recap left this writer unsure of its value

The writer tested Apple’s Siri Recap on an Apple Watch Ultra 4 for two weeks, comparing it with Amazon Bee Frontier, which also listens to conversations and creates AI summaries. Siri Recap sometimes treated audio from videos as if the events had happened to the writer, and its summaries often focused on facts, dates and figures rather than the details the writer found most meaningful. It helped with some work tasks, including remembering meeting topics and linking interview recaps to Voice Memos files. The writer found few other uses for it and felt its privacy-focused approach was preferable to Bee, whose stored transcripts and summaries led the writer to delete all Bee data.

Groundcover creates shared investigation workspace for AI agents and engineers

Groundcover launched Agent Investigations, a shared workspace where engineers and AI agents can investigate production incidents together. Investigations can start automatically when a monitor issue opens, or engineers can start them through Groundcover or Slack. An Investigator agent tests questions and hypotheses against logs, metrics, traces and Kubernetes data, while a Moderator agent checks the evidence before accepting or rejecting a conclusion. Outside agents, including Anthropic’s Claude Code and OpenAI’s Codex, can also join through Groundcover’s Model Context Protocol server. The service is available to Groundcover customers with AI enabled on their backends, and the company will demonstrate it at AWS re:Invent in Las Vegas from November 30 to December 4.

Sources

NOTE:

This news brief was generated using AI technology (including, but not limited to, Google Gemini API, Llama, Grok, and Mistral) from aggregated news articles, with minimal to no human editing/review. It is provided for informational purposes only and may contain inaccuracies or biases. This is not financial, investment, or professional advice. If you have any questions or concerns, please verify all information with the linked original articles in the Sources section below.

OpenAI GPT-6 Cyber Java4Shell OpenJDK Base64 decoder vulnerability Remote code execution Cybersecurity Vulnerability disclosure Oracle security patch CERT-EU CISA Anthropic IPO AI investment Astera Labs AI infrastructure Token-based pricing AI operating costs Groundcover Agent Investigations AI incident response AI translation Siri Recap

Comments

Loading...