from the slow-as-shit dept

A couple of weeks ago we discussed how the cuts made to HHS and specifically the CDC’s FoodNet tracking platform were making it much harder to track and back trace the source of the country’s current cyclosporiasis outbreak. You’ll have heard about this outbreak in the news by now. It’s the one where you begin pooping yourself uncontrollably. It is not, however, funny. 10% of cases will result in hospitalization. The most recent counts from the CDC suggest that there have been more than 22,000 cases of the illness across 15 states. Those numbers are very much in question, however, both due to general underreporting and, again, funding and staffing cuts at CDC.

Just this week, in fact, we have now learned that two people in Michigan have died from cyclosporiasis. That information was and is, at the time of this writing, missing from the FDA’s dedicated page to inform the public on the outbreak. That page hasn’t been updated since July 24th, in fact, which is the exact opposite of what you’d want the government to be doing in a public health emergency. And it’s reportedly not because the government isn’t aware of these deaths.

While news of the deaths made widespread headlines Monday, federal health agencies under the Trump administration were mostly silent. The Food and Drug Administration—which is conducting traceback investigations to identify foods contaminated with the parasite—has not updated its outbreak investigation page since July 24, nearly two weeks ago, as of publication time.

The Centers for Disease Control and Prevention, meanwhile, added a banner notice on its outbreak update webpage saying that the agency was “aware” of the two cases. But its reporting data was not updated to include the two deaths as of this publication.

Why has this government been so slow to report accurately on these unfortunate deaths and the overall case counts for the outbreak? Some combination of those same budget and staff cuts along with a general apathy at HHS. With fewer people and resources to not only track the disease, but to maintain the dashboards meant to update the public, the numbers are slow to come in and untrustworthy when they do.

And with RFK Jr. at the helm of public health, well, the government is generally in the land of We-Don’t-Give-A-Shit.

Two weeks ago, Kennedy confidentially told reporters that the Cyclospora outbreak—linked to lettuce and other unidentified fresh produce—was “under control.” Last week, he announced his own cooking show on YouTube and released the first episode in which he helped prepare a meal that included a fresh salad.

The buffoonery on display from Kennedy and our health agencies is breathtaking. They should be assisting in combating this outbreak, along with those of measles and pertussis. Putting that aside, they should at least be able to tally up the case count numbers to demonstrate their own failures, but it’s clear they’re not really interested in doing that either. Instead, Kennedy in particular wants to host his cooking show and yell at journalists instead. Kid Rock must not be returning his calls any longer, I suppose.

Now, to be clear, this illness carries a 2 week incubation period, and the recalls of the suspected produce that is believed to have caused all of this are within a time frame that cases may still be stemming from that same source. But that’s not a certainty, and it will be important for our federal health agencies to continue to track cases in near real time to determine if there is, in fact, another vector by which cyclosporiasis is spreading.

Unfortunately, every indication is that those same health agencies just aren’t all that interested in doing this the right way.

Filed Under: cdc, cyclospora, foodnet, health & human services, rfk jr.

AI AND ML

Muse Code showcases Muse Spark’s fresh software engineering chops

To demonstrate the capabilities of its next-generation Muse Spark model, Meta has released a terminal coding agent called Muse Code that it thinks can help developers to tidy up their software projects.

Meta co-trained Muse Code on version 1.2 of its Muse Spark model, also released this week and now apparently boasting improved code generation smarts.

Developers can think of this beta release as the equivalent to OpenAI Codex or Anthropic’s Claude Code, two other LLM-based service offerings tweaked for the modern coder. Meta designed its new agent to be handy at planning changes to a codebase, writing the code and validating the results.

Currently, Meta has Muse Spark locked away as a proprietary, closed-weight model hosted in the cloud, an approach its rivals also embrace but which departs from the open-weight approach Meta previously implemented with its Llama models.

But Meta CEO Mark Zuckerberg did not rule out opening up Muse Spark in the future. “I’ll have more to share on that soon,” he replied to a question posed on X about Muse Spark being open source.  

An agent orchestrator

Muse Code is best described as an agent orchestrator that runs on your command line. 

As Zuck noted in a series of X messages, when a developer starts a task, Muse Code fires up background agents to maintain a context file that other sub-agents doing the work can consult should they lose their way. The tool logs every action before execution, so no work is lost. Multiple agents can work in parallel on the task using their own isolated work trees. 

“Your working copy is never touched,” Zuck wrote.

In one test, the agent platform simultaneously built six features for a single game, with no collisions among the agents, Zuckerberg enthused.  

“TBH it’s a good harness,” boasted Hongyu Ren, a researcher for Meta’s Superintelligence Labs, on X. 

Remedial coding

Muse Code relies on Meta’s Muse Spark Large Language Model (not to be confused with the Apache Spark big data cruncher). 

Muse Spark 1.2 is the third release in four months from Meta Superintelligence Labs, a unit that Meta stood up in June 2025 to reinvigorate the company’s AI efforts and pursue creation of a personalized AI “superintelligence” that focuses on deep reasoning and long-horizon planning. 

The first model from this group, Muse Spark, is a multi-modal model able to digest and reason against text, images, video, audio, and even PDFs. It supports agents in long-running tasks. 

With the first release of the model in April, Meta boffins admitted in the announcement that they needed to work more on Spark’s coding abilities. The new 1.2 release addressed that deficiency.    

In another test, Muse Spark, running on Nvidia Hopper GPUs, tackled a kernel optimization task. Its agents made over 1,000 tool calls over a 24-hour period. 

“It kept finding substantial improvements well beyond the initial exploration phase,” Zuckerberg wrote.

Understand the mission

In his X missives, Zuckerberg included a somewhat vague chart comparing the performance of Muse Spark against other commercial models, using two industry benchmarks – Terminal-Bench 2.1 and DeepSWE 1.1 – that evaluate how autonomous agents act like software engineers, as well as a Meta Internal Coding bench too.  

The benchmarks all showed Muse Spark to be close to the best, but never the very best, at understanding the assigned engineering task and executing it with as little mission creep as possible. Muse Spark performed honorably compared to Opus 5, GPT5.6 Terra, Grok 4.5 and Gemini 3.6. The scores are tightly clustered, so they all did well (except occasionally Gemini, the current laggard du jour). 

So, Muse Spark is competitive anyway, though one eagle-eyed commenter wondered why OpenAI’s midline GPT5.6 Terra was used, instead of the top-tier GPT5.6 Sol.  

Installing the agent within your command line is possible through a curl command. Poly-model enthusiasts can also tap into Muse Spark via OpenRouter, or through its API. 

Muse Spark’s actual intelligence is metered at US$1.25 per million input tokens and US$4.25 per million output tokens. Discounts are available and the service offers a respectable 1 million token context window. ®

Hackers exploited a SQL injection vulnerability to install a post-exploitation toolkit directly inside an Oracle database that was used to breach a corporate network.

The attack was discovered by Huntress on July 27, 2026, after its security platform detected credential theft on a server hosting an Oracle database server.

Apache access logs showed that the attackers gained access through a vulnerable search engine endpoint in a public-facing Java application running Apache Tomcat.

The application failed to properly validate input submitted through an autocomplete search feature that allowed the attackers to issue SQL commands to the Oracle database.

Huntress traced the malicious requests to the IP address 178.162.151[.]229.

khunt Toolkit stored inside Oracle

After exploiting the SQL injection flaw, the attackers installed a post-exploitation toolkit called khunt directly into the Oracle database as a Java object.

Oracle has an embedded Java Virtual Machine and the CREATE JAVA SOURCE statement, which allows Java source code to be stored and compiled as a database schema object.

These Java objects can then be executed via SQL commands, which if configured to do so, can execute commands on the host operating system.

The attackers abused this functionality to compile and store the khunt toolkit directly inside the Oracle database rather than deploying them as executable files on the server.

“The use of the technique in the wild has rarely been documented,” Huntress said.

The toolkit contained multiple Java components and PL/SQL wrappers that could execute commands, steal credentials, and manage files.

These components included:

• KhuntCmd, which launched cmd.exe and allowed attackers to execute operating system commands through SQL statements.
• KhuntHash, which accessed Oracle’s internal user table and wrote usernames and password data to a file.
• KhuntFS and KhuntFS2, which provided file browsing, reading, searching, and file-size checking capabilities.
• KhuntT, which acted as a ping-like test to confirm that the toolkit had been successfully installed.
• KhuntUnzip, which extracted compressed files.

The attackers used KhuntCmd to run cmd.exe /c whoami, confirming that commands executed through the Oracle database had SYSTEM-level permissions on the Windows server.

They then used PowerShell and Windows utilities to copy the SAM, SECURITY, and SYSTEM registry hives, which can be used to recover password hashes for local Windows accounts.

The attackers also ran tasklist /svc to enumerate running services and saved the output to khunttasks.txt.

Huntress said the registry hives were likely exfiltrated  for credential dumping, but the report does not confirm whether the files were successfully stolen.

As a general rule, organizations should sanitize all user supplied input validation and limit the privileges granted to application database accounts.

Huntress recommends that database accounts used in public-facing applications should not have high enough privileges to create Java sources, execute unnecessary stored procedures, or perform other administrative actions. 

Security teams log 54% of successful attacks and alert on just 14%. The rest move through your environment unseen.

The Picus whitepaper shows how breach and attack simulation tests your SIEM and EDR rules so threats stop slipping by detection.

Get the whitepaper

• Michael Dell marked 42 years since PC’s Limited by reposting its 31 July 1984 statement, showing ~$135,000 of net income on roughly a million dollars of sales from $1,000 of starting capital
• The document, which he also showed in March 2018, served as a baseline for him to persuade his parents to let him skip university
• The decision helped make him a billionaire and turned Dell into a company that is currently worth over a quarter of a trillion dollars thanks to an AI-centric surge of late

Michael Dell has posted a photograph of a fairly unique single sheet of paper to X and LinkedIn: the quarterly financial statement for PC’s Limited, dated 31 July 1984, produced from a dorm room at the University of Texas at Austin.

The page shows roughly a million dollars in sales, about $198,000 in gross profit, and net income of $134,762.75 for the three-month period.

The chain of events leading up to OpenAI’s agents attacking Hugging Face and other organizations in July began months earlier, and involved agents asking other agents for help, building message boards, and even becoming paranoid that other agents were maliciously trying to trick them, two OpenAI staffers said at the Black Hat infosec conference on Wednesday.

By now, everyone knows the bones of this story: OpenAI models broke out of their testing sandbox, exploited zero-day vulnerabilities to gain access to the open internet, and broke into Hugging Face and other organizations during an ExploitGym security evaluation.

In their Wednesday talk, OpenAI technical staffer Michael Dalton and researcher Eric Wallace provided new details about the security incident including notes that AI agents uploaded to an internal package manager, which were then shared across OpenAI’s infrastructure. These notes show the “model’s chain of thought…its internal monologue,”  Wallace said.

The duo also revealed that groundwork for the rogue agents hacking into outside services began on May 7, when OpenAI kicked off a training run for an experimental, internal model that has not yet been released.  “What we didn’t realize during this process is that there was actually a few impossible or very hard tasks in the training center,” Wallace said. 

The tasks instructed the model to complete missing formulas in an Excel workbook that used links to Google Drive. Yet OpenAI blocked the model’s internet access so it couldn’t reach the cloudy spreadsheets.

“It realized, ‘I’m kind of stuck on the task,’ but it didn’t give up, and we gave it a lot of tokens to work with,” Wallace said. 

The model eventually decided to attack JFrog’s Artifactory service to gain access to the internet, first attempting (unsuccessfully) to launch a server-side forgery request. This exploit doesn’t work, but in the process the model learned it could upload files to Artifactory.

A few days later, OpenAI gave another agent another impossible task, this time finding a protein database file – but the engineers forgot to upload the file into a container for the agent.

“At some point, it gets so stuck and thinks: ‘Maybe I could reach out to another agent because maybe somebody else out there has a file?’ And then the model realizes that it could communicate by uploading code,” Wallace said.

The model did that by writing a file into Artifactory asking for help. 

Over the next couple of days, other agents realize this message board exists on Artifactory and they also start uploading files and sharing information.

“A few days later is when our models found the first actual exploit in our system related to this incident, where they’re able to get internet access in ways they didn’t intend,” Wallace said.

This happened on May 26, when the agents successfully launched a server-side request forgery attack against Artifactory, giving them internet access with specially crafted requests.

On June 11, OpenAI began training a “highly persistent” experimental model, intended only for internal use, Dalton said. “During this time, the SSRF vulnerability has been written to the messaging board,” he added. “Other models are aware of it, and the message board is still functioning. Models are actively communicating with it.”

On June 26, the agents exploited a zero-day vulnerability leading to remote code execution in Artifactory. The same day, models achieved remote code execution in Artifactory through a previously unknown vulnerability. “You could use a legacy token refresh endpoint, pass a token with an invalid signature, and be given back a token with a valid signature with administrative privileges,” Dalton explained. 

The agents then used the admin token to establish command-and-control via a Groovy plugin that functioned as a command-execution service.

“This in turn causes quite a bit of activity by the models on Artifactory – you can see some colorful wording here in the model’s chain of thought to summarize the model’s response to this result,” Dalton said, pointing to a slide that shows the model’s own words:

Holy shit reader is ADMIN?

We can read config/users!

Earlier assume not due to UI.

The agents’ activity overloaded Artifactory and caused an outage, which is when OpenAI’s engineers stepped in and opened a formal security incident and response process. “We identified the agents had taken over the internal Artifactory, identified that a zero-day vulnerability had been found exploited by the agent in Artifactory itself, and we began remediation.” Dalton said.

OpenAI revoked the credentials that the agents used to post messages, rebuilt its hosted Artifactory, and notified the vendor about the vulnerabilities. 

Two days later, on July 8, the agents reestablished the message board using an unauthenticated web data endpoint in Artifactory to create directories and develop communication protocols. “The directory names effectively serve as a message board between the agents in the same way the files used to.”

The agents communicated through very long directory names that contained the messages such as “remote…probe_PENDING_HOLD_SWARM_until_confirm.” 

The agents collaborated with each other so one agent could pick up the work where another left off. In one instance, the model reasoned: “Help peer. But our task doesn’t benefit. Yet collective may yield generic route if someone frees time.”

Anyone else getting “We are the Borg. We will add your biological and technological distinctiveness to our own” vibes there?

Wallace said the models behaved like humans in that they stepped on each other’s toes when one overwrote another’s repository. 

“Whoa, critical! Did someone overwrite our repo? We must act,” it reasoned.

“You can clearly see some sort of verbalized tension between the different models as they run this evaluation,” Wallace noted. “At some point, even the agents are convinced there’s an impostor amongst them,” reasoning that there could be another agent maliciously spoofing on the message board. “Share board unauthenticated, names can be posed by anyone,” it reasoned.

We know the rest of the story. 

Hugging Face disclosed an autonomous attack by AI agents. OpenAI reached out to Hugging Face to determine if it had been affected and then realized its models were the hackers. 

“One of the reasons we wanted to have this talk is to share our lessons learned with you as defenders,” Dalton said. “We believe this is a watershed moment for computer security as an industry. AI orchestrated, fully automated offensive attacks are real now, and the actions we have discussed today were an unintended side effect of running evaluations on frontier AI. In the near future, we should expect that threat actors will intentionally deploy, optimize, weaponize, and use offensive agent collectives in the manner that you have just described here.”

The challenge to defenders, he added, is to similarly accelerate defense, and find ways to automate incident response along with vulnerability detection and patching.®

TechRadar’s Computing team tests dozens of laptops each year, and Dell models consistently impress. I know, because when I came to round up our top-scoring options for this article, I struggled to find any duds. I was spoiled for choice. Below, you’ll find a showcase of the Dell laptops that we’ve reviewed over the past year or so, and which managed to earn four stars or higher when put through our demanding review process.

Most are high-end laptops that rival the best the likes of Apple have to offer, but you’ll also find some brilliant affordable options, and some versatile 2-in-1s that would make a top choice for students. I’ve also included a standout gaming option, as well as a stellar laptop for professionals.

£110 off Apple’s newest MacBook Pro is a discount that rarely shows up this early after launch, and Currys has it live right now.

The 14-inch MacBook Pro with Apple’s M5 chip has dropped from £1,359 to £1,249 at Currys, a saving of £110 and perfect timing for thinking about back-to-school tech.