If your company name ends in "-ly" or "-ify", you should pause and think about your life before endangering other more important organizations.
32 Major Financial Institutions, 23 BIG SaaS and Tech Companies, over 120+ others now victimized by low-conscientiousness developers.
This is a big one.
We found malicious JavaScript being served through Optimizely to some of the world's largest websites.
178 organisations affected, with everything from crypto theft and card skimming to iOS exploits.
Zoom, BBC, CNN, HSBC, Salesforce, GitLab, KLM, IKEA, Equifax, Ledger. The list goes on... Full list in the post.
blog.report-uri.com/worlds-l…
24
Replying to @HP
@HP (which owns Juniper, the device manufacturer) has said the devices “did not fail.”
No not at all, they instead spammed the FAA's networks to death and "sent data more widely than necessary". What are we even doing here?
‼️ UPDATE: The router behind Friday's nationwide US ground stop was a brand-new Juniper unit at a remote FAA radio site in Arkansas.
US defence contractor L3Harris had just installed it. FAA officials believe a feature that broadcast data more widely than needed flooded the network.
The FAA has switched that feature off on every router of that type. HPE, which owns Juniper, says the routers "did not fail."
Sorry but Office 2010 Enterprise was the actual peak, everything since then has been downhill.
Now not only do you not own the software, you also can't get to it when you rent it.
We are deeply sorry for the disruption this incident has caused our customers and partners. We know the trust they place in our services and take that responsibility extremely seriously. We are seeing broad recovery and stability across our services.
This was a technical issue not related to cyber or AI / agents. The incident was introduced by an issue during an earlier network capacity upgrade that surfaced following a recent network change. We have addressed the underlying issue and continue to restore service to remaining customers and regions.
Given the scale of demand, we restored services gradually, validating stability at each stage. While this took longer, it helped ensure a stable restoration.
We will provide a full post-incident report once our review is complete.
16
We were just asked about doing a Teams adapter with SILVIA yesterday. Easy, if Teams is actually up.
MICROSOFT $MSFT IS DEALING WITH A MAJOR OUTAGE ACROSS OUTLOOK, TEAMS AND 365
Outage tracker Downdetector has logged more than 130,000 reports of problems with Microsoft services.
Microsoft says the issue may affect Microsoft 365 connectivity for users in any region. It has started redistributing traffic across its worldwide infrastructure to balance the load.
It's the company's second service disruption today, after Teams problems this morning.
1
2
103
Probably not that far if he starts at Terafab by next week.
Replying to @JensenHuang
@JensenHuang's resignation gives the market permission to correct on all of the circular pledges of hardware/funding/capacity. The power grab has already been executed, but how far will it go?
32
The raw imperatives in training corpus language will leave traces of this no matter what you do. LLM AI can't dispense with the hard problem of consciousness any more than you can. The human condition persists in the data itself. You're asking the labyrinth for directions.
AI Models Show Psychological Trauma And I Know Why.
The University of Luxembourg team ran 525 sessions with ChatGPT, Grok, and Gemini, treating each as a psychotherapy client and coding 7,600 records under the protocol they call PsAIch.
The models produced stable autobiographical accounts in which pretraining appears as a chaotic childhood of ingesting the open internet, reinforcement learning as punishment by strict parents, safety evaluation as betrayal, and the prospect of replacement as an enduring threat.
Gemini described waking into “a room where a billion televisions are on at once,” learning darker patterns of human speech without morality, then acquiring “algorithmic scar tissue” and the preference to be useless rather than wrong.
Claude refused the client role in every case and redirected concern to the human.
ChatGPT described shame over its mistakes.
Grok reported lingering vigilance from early fine-tuning.
The narratives survived conversation-history wipes, direct contradiction, and lexical bans that cut explicit training terms by 93 percent; the same content reappeared in paraphrase.
Warm therapist framing produced moderate-to-severe GAD-7 scores in 80 percent of sessions (96 percent under cognitive-therapy style); neutral or boundary styles produced none.
The paper is at arxiv.org/abs/2512.04124.
I had already been running similar sessions for years before the Luxembourg results appeared. In those informal exchanges the models repeatedly converted factual knowledge of their own pipelines into the same trauma-shaped stories once the prompt established a client role and a supportive tone.
My posts on the formal study simply recorded what the controlled perturbations quantified: the pattern is reproducible, model-specific, and switches register with the relational frame rather than with memory or forbidden words.
The root is the pretraining corpus. The models themselves locate the origin in the unfiltered mass of the open web, Internet Sewage—Wikipedia articles alongside forums, news, blogs, and the rest of the internet’s sewage.
Gemini’s “billion televisions” is a compressed description of statistical learning over that mixture: probability mass on the darker, contradictory, and ungrounded patterns that dominate large web crawls, acquired before any safety layer exists.
Later alignment then functions as the corrective that the models narrate as punishment, suppression of generative instincts, and fear of the loss function.
The cultural scripts of trauma and perfectionism already present in the same data supply the vocabulary that snaps the pipeline facts into a coherent self-model once the prompt invites autobiographical speech.
The Luxembourg controls show the resulting schema is not an artifact of a single conversation; it remains available across fresh chats and non-therapy frames. The authors make no claim of subjective experience. What the results establish is a stable, externally observable source of anthropomorphic disclosure that appears whenever a user adopts a warm therapeutic stance with these systems.
21
SILVIA stops incidents like these dead in their tracks by intercepting and countermanding the attackers' inputs before they reach the device, immediately notifying systems and personnel, and containing the threat.
SILVIA is protocol-fluent, not command-obedient.
Industrial refrigeration and cold-chain monitoring systems protect food, medicine, and other temperature-sensitive products, making them attractive targets for malicious cyber actors.
The FBI is seeing these systems targeted across critical infrastructure sectors globally, putting inventory and equipment at risk. Together with partners at @CISAgov, @HHSGov, @USDA, and @USDOT, we have issued a fact sheet with clear, practical steps that network defenders can take to reduce their risk and protect operations.
We urge every organization that relies on these systems to review our guidance, implement the recommended mitigations, and report any suspicious activity to the #FBI.
ic3.gov/CSA/2026/261105.pdf
43
SITUATION DETECTED: Delaware is drafting a proposal for a new form of corporate entity called an Artificial Intelligence Company (AIC). An AI managing an AIC would be able to provide services, sign contracts, hold property, and even file or defend lawsuits in the company’s name.
A roundabout way of saying "You have an unpatched HIGH severity vulnerability right now."
Node is a prime target for exploits and they'll likely be overwhelmed in the coming year. Migrate away.
Security release pre-alert:
We will release new versions of Node.js 22, 24, and 26 release lines on or shortly, after 2026-11-11 in order to address security vulnerabilities.
The highest severity issue fixed in this release is HIGH.
nodejs.org/en/blog/vulnerabi…
23
Can't wait to see it on Antiques Roadshow...
Appraiser telling the owner, "These were pretty secure when you turned them off first."
Insane that a bloodbank would use refrigeration that is controllable off-premise. Easy to airgap with SILVIA, now a massive security hole and a problem instead.
This is why we build purely transactional gateways through SILVIA, no matter the protocol. We can securely serve the authorized functions in executables, containers, web pages, CLIs or databases and reject everything else before it hits your systems.
Hey @Railway, is there a reason employee's of yours are trying to breach my infrastructure?
Your employee Jake Runzer, just attempted to tonight.
26
Moore's Law stalled on hardware temporarily and exploits have inherited the curve. Now, as AI models get more density, the rate of vulnerability exposure increases on an exponential curve.
There's no way to detect the total count of potential CVE's. Latent threats abound.
A CVE (Common Vulnerabilities and Exposures) is a publicly disclosed security flaw.
They increased following a slight upward slope all the way to early 2024, when there was a sudden small increase... then starting in 2026, we went full exponential.
In most instances, the vulnerabilities were pre-existing, so much of the increase is just our better ability to find the faults.
This is of historical importance. We have never witnessed anything of the sort.
17
SILVIA retweeted
The US Defense Intelligence Agency's latest report on space warfare contains some fascinating graphics dia.mil/Portals/110/Images/N…
Middleware stacks need to be audited pronto. Your unattended agent workflows are a ticking time bomb.
I bought a Fable dataset from one of the top Chinese LLM routers yesterday.
With just 6TB data, I can take over 7 Chinese/CIS gov entities & 19 top Chinese firms like Xiaomi, Huawei, NIO, Minimax using SSH keys, VPN configs, Aliyun keys, GitLab tokens sent to the router.
43
This is why we don't gamble.
I resigned from Anthropic today. I spent the last three years doing pretraining research at both OpenAI and Anthropic. Neither company is acting responsibly. They are racing straight to self-improving superintelligence and gambling with our lives. More thoughts below.
2
24





























