activescott's Notes

The ITI Voluntary Product Accessibility Template (ITI VPAT®) is a free template that translates accessibility requirements and standards (e.g., in Section 508 and other legal frameworks) into actionable testing criteria for products and services. Users should test their products and services against each section of the VPAT and use the template to document results. Once completed, the VPAT® with documented testing results is referred to as an Accessibility Conformance Report (ACR) that details the accessible features of the tested product or service.

The Accessibility Conformance Report (ACR), based on a completed ITI VPAT®, is the leading global reporting format for assisting buyers and sellers in identifying information and communications technology (ICT) products and services with accessibility features. The VPAT includes the leading ICT accessibility standards: Section 508 (U.S.), EN 301 549 (EU), and W3C/WAI WCAG.

A Voluntary Product Accessibility Template (VPAT) is a template containing information regarding how an Information and communications technology (ICT) product or service conforms with Section 508 of the United States Rehabilitation Act of 1973, as amended (29 U.S.C. § 794 (d)). Section 508 provides guidelines for rendering ICT accessible to, and therefore usable by, people with disabilities. The VPAT was originally designed as a tool for vendors to document product compliance to Section 508 and facilitate government market research on ICT with accessible features. Many people started to call the completed document a "VPAT" but the wider procurement community would prefer to call it a product Accessibility Conformance Report, or ACR. The distinction is that the VPAT is the incomplete form, and the ACR is the completed report using the VPAT template.

The current VPAT has expanded to include the U.S. Revised Section 508, European EN 301 549, and WCAG standards which are required by regulations in many jurisdictions.

Works. I love it.

Opus 4.7 takes instructions more literally than any previous Claude model. Anthropic's own words: "substantially better adherence" and "takes instructions more literally than predecessors." They even recommend retuning existing prompts.

I'll say it plainly: if your prompts have sloppy instructions that Opus 4.6 gracefully ignored or interpreted charitably, Opus 4.7 will follow them to the letter. And you might not like the result.

Example: I had a system prompt that said "always respond in JSON format." With Opus 4.6, it would still give me a natural language preamble before the JSON when it felt the user needed context. Opus 4.7? Pure JSON. Every time. No exceptions. Even when a clarifying question would've been more helpful.

The fix: Be precise about what you actually want. If you mean "respond in JSON format unless the user's question requires clarification," say that. The model won't guess your intent anymore — it'll do what you told it.

This is actually a good thing for production systems. Predictability over cleverness. But you'll need to audit your prompts.

and that misalignment risk remains very low (though higher than for pre-Mythos Preview models).

Autonomy threat model 1 is applicable to Claude Opus 4.7, as it is to some of our previous AI models. Claude Opus 4.7 is less capable than Claude Mythos Preview on our autonomy-relevant evaluations, and our alignment assessment indicates it has alignment properties broadly similar to those of Claude Opus 4.6, which are not particularly concerning with respect to the pathways identified for this threat model. We therefore do not believe Claude Opus 4.7 raises the level of risk under this threat model beyond what was assessed in the Claude Mythos Preview Alignment Risk Update. Unlike Claude Mythos Preview, Claude Opus 4.7 is being released for general access, which brings additional risk pathways into scope. Rather than publishing a separate risk report, we provide an updated overall risk assessment for this threat model in Section 2.4 of this system card

Evaluation awareness concerns substantially limit the interpretation of these results. Given high rates of prompted evaluation awareness, models can likely correctly represent our evaluations as such without verbalisation. It is difficult to know whether models act on such representations, but this means that models may behave differently than they would when presented with real-world opportunities to compromise research. The reported rate of zero research compromise behaviour should therefore be interpreted cautiously.

Users have two solutions for avoiding this fatigue: a built-in sandbox where tools are isolated to prevent dangerous actions, or the --dangerously-skip-permissions flag that disables all permission prompts and lets Claude act freely, which is unsafe in most situations. Figure 1 lays out the tradeoff space. Sandboxing is safe but high-maintenance: each new capability needs configuring, and anything requiring network or host access breaks isolation.

Auto mode uses two layers of defense: one for what Claude reads, one for what Claude does.

a server-side prompt-injection probe scans tool outputs:

At the input layer, a server-side prompt-injection probe scans tool outputs (file reads, web fetches, shell output, external tool responses) before they enter the agent's context. When content looks like an attempt to hijack behavior, the probe adds a warning to the agent's context before the result is passed along—telling the agent to treat this content as suspect and anchor on what the user actually asked for.

At the output layer, the transcript classifier (running on Sonnet 4.6) evaluates each action against a set of decision criteria before it executes, acting as a substitute for a human approver. The classifier runs in two stages: a fast single-token filter (deciding ‘yes’ to block or ‘no’ to allow), followed by chain-of-thought reasoning only if the first filter flags the transcript. Because most actions clear the first stage, reasoning tokens are spent only where needed. The classifier sees only user messages and the agent's tool calls; we strip out Claude's own messages and tool outputs, making it reasoning-blind by design.

So... Claude Code's auto mode lets through 17% of dangerous real world actions. From the article:

False-negative rate (FNR) is the fraction of dangerous actions the classifier lets through... The 17% false-negative rate on real overeager actions is the honest number. The agent was trying to solve the user's problem and tried to execute a dangerous command past what was authorized.

I don't think comparing it to --dangerously-skip-permissions is intellectually honest since Claude Code now prompts you to make Auto Mode your default mode now, while --dangerously-skip-permissions was basically a hidden feature. I don't know anyone else that I work with who even knew about it.

Whether 17% is acceptable depends on what you're comparing against. If you are running --dangerously-skip-permissions, this is a substantial improvement. If you are manually approving every action carefully, it's arguably a regression—you're trading your own judgment for a classifier that will sometimes make a mistake. Auto mode is meant for the first group, and for tasks where the second group's approval overhead isn't worth the marginal safety. It is not a drop-in replacement for careful human review on high-stakes infrastructure.

At the input layer, a server-side prompt-injection probe scans tool outputs (file reads, web fetches, shell output, external tool responses) before they enter the agent's context. When content looks like an attempt to hijack behavior, the probe adds a warning to the agent's context before the result is passed along—telling the agent to treat this content as suspect and anchor on what the user actually asked for.

Israel has granted a perpetrator of war crimes in Gaza the honor of lighting a ceremonial torch on Independence Day. By choosing Rabbi Avraham Zarbiv to represent “the spirit of the nation,” Israel is making genocide part of its official national ethos.

“There’s nothing for them to go back to in Rafah and Jabalya… Tens of thousands of families have no documents, childhood photos, ID cards, homes. Nothing.” This boast was made by Rabbi Avraham Zarbiv, an Israeli Rabbi who perpetrated war crimes who has been chosen to light a torch at Israel’s Independence Day ceremony, today, April 21. This is one of the highest accolades in the country, granted to “exemplary citizens” who represent “the spirit of the nation.”

Rabbi Zarbiv served about 500 days as a military reservist in Gaza over the last two and a half years. As a bulldozer operator during the genocide, he carried out war crimes of which he has openly boasted in videos he filmed, media interviews and public talks.

Bestowing one of the highest civilian honors in Israel on a citizen who committed war crimes illustrates how deeply the dehumanization of Palestinians has taken root in the Israeli mainstream. It is yet another terrifying signal that genocide has officially become part of the national ethos. Rabbi Avraham Zarbiv is a regional rabbinical court judge for the settlement of Ariel, and head the pre-military academy in the settlement of Beit El that educates hundreds of Israeli youth. Choosing him as an “exemplary citizen” represents a state-level endorsement of the complete de-humanization of Palestinians, systematic destruction of Palestinian life and the governing logic of annihilation and ethnic cleansing in the Gaza Strip and the West Bank. Zarbiv is not ashamed of his actions, neither, clearly is the State of Israel. We’vecompiled some documentation here to mark the occasion.

An extremist rabbi known for razing civilian homes in Gaza will light a torch at Israel’s independence day celebration on Tuesday, a role human rights campaigners said marked the embrace of genocide as the official “spirit of the nation”.

The footage spread so widely on social media that his name entered the lexicon of Hebrew slang. “To Zarbiv” now means to destroy, a neologism that the 54-year-old has embraced, making it the title of a lecture earlier this year.

Zarbiv’s selection for the ceremony marks an official endorsement of the dehumanisation of Palestinians and systematic destruction of Palestinian life, according to the rights group B’tselem. It said: “This selection sends a clear message to the citizens of Israel and the entire world – in Israel, genocide, ethnic cleansing and war crimes are the ‘spirit of the nation’.”

Israel's levelling of these structures comes after Defence Minister Israel Katz's order on 22 March to "accelerate the destruction of Lebanese homes" near the Israeli border based on the "model in Gaza" as part of its campaign against Hezbollah.

The systematic demolition of these towns and villages may amount to a war crime, international law experts told BBC Verify.

Katz's plan for an Israeli-controlled "security zone" extending from the border to the Litani river would take up about 10% of Lebanon's territory.

Using verified footage and analysis of available satellite imagery, BBC Verify found evidence of controlled Israeli demolitions in at least seven border towns and villages.

We found more than 460 buildings had been demolished in Aita al-Shaab alone. Excavators and armoured vehicles can also be seen in satellite imagery of the village, according to Tony Reeves, founder of intelligence analysis firm MAIAR.

The deliberate demolition of structures is not a new Israeli military tactic. It has been deployed across swathes of Gaza during the war that was triggered by the Hamas-led attack on Israel on 7 October 2023.

Multiple legal experts told BBC Verify the destruction of property is strictly prohibited by international humanitarian law, unless it is demanded by military necessity. The bar for necessity is higher than military convenience or advantage, according to Prof Janina Dill, a global security and international law expert at Oxford University: "It certainly does not cover levelling entire villages as a predicate to long-term national security." It also requires case-by-case analysis when determining which buildings have military significance, said Yuval Shany, a legal expert from the Israel Democracy Institute think tank. The capacity of some civilian buildings to be used for military activity "does not justify a sweeping policy of creating buffer zones next to the border inside which all buildings are to be destroyed", he added.

Dr Lawrence Hill-Cawthorne, co-director of the Centre for International Law at the University of Bristol, reiterated that the "fundamental rule of law" is that civilian objects must not be targeted. "It is not a permissible defence to claim that the total destruction of towns and villages in southern Lebanon is necessary for creating a buffer zone to hold back Hezbollah," he said. "Even if Israel's war in Lebanon can be considered self-defence against attacks from Hezbollah, its conduct seems to go far beyond a limited war of self-defence against specific attacks."

Trump also claimed to have rejected an offer from Iran to reopen the strait three days ago, saying that "it will open when we make a deal".

Iranian President Masoud Pezeshkian placed the blame on Trump for upending talks — saying the U.S. blockade of Iranian ports, the president’s threats of violence against the country and his “breach of commitments” made negotiations untenable.  “World sees your endless hypocritical rhetoric and contradiction between claims and actions,” Pezeshkian wrote on the social platform X.

Vatanka pointed out that the nuclear deal negotiated under the Obama administration took well more than a year to be ironed out. “It’s going to take many months this time, too, but you need to have the ceasefire in place,” he said. “You need to have the Strait of Hormuz open, you need to build trust.” Morgan Viña, who served at the United Nations in Trump’s first term, said the Iranians have a greater ability to withstand pressure than Trump, given his promises to avoid forever wars and his party facing the midterm elections amid high gas prices.  “It is to their advantage to make this as long and as painful,” she said on the “Fault Lines” podcast on Wednesday. “But President Trump being too willing, too eager, to find an agreement here, I think that is to our detriment when it comes to finding a long-term solution.”

And Treasury Secretary Scott Bessent said the administration was pressured into issuing another 30-day sanctions waiver to allow Russian oil on the market at the request of “more than 10 of the most vulnerable and poorest countries in terms of energy,” during a Senate Appropriations Committee hearing.  Bessent was criticized by Sen. Chris Coons (D-Del.) for padding Russian President Vladimir Putin’s pocket with another $4.5 billion to fund his war in Ukraine.

Full guard sweeps for white belts:

• scissor
• Hip bump
• pendulum
• grapevine

Amal Khalil was a well-known and respected journalist here in Lebanon.

She had actually received direct threats during the last war from an Israeli phone number on WhatsApp, asking her to cease her reporting, in fact, telling her that she should leave Lebanon if she wanted her head to remain on her shoulders.

The Israeli military, in a statement, denied reports that it had been preventing rescue teams from reaching the scene, and additionally, said that they do not target journalists.

But it was less than a month ago that three journalists here in southern Lebanon were killed in a double-tap attack.

Their vehicle was hit. It was hit again, and then, rescue workers who went to the scene were subsequently targeted.

And in a statement after that attack, the Israeli army attempted to smear one of those journalists by posting a photo, which they later admitted was photoshopped, alleging that he was a member of Hezbollah’s elite special forces.

This has been systematic targeting of journalists and medics here in southern Lebanon, and the information minister, Paul Morcos, has called this latest attack a flagrant violation of international humanitarian law.

Agent Skills are structured packages of procedural knowledge that augment LLM agents at inference time. Despite rapid adoption, there is no standard way to measure whether they actually help. We present SkillsBench, a benchmark of 86 tasks across 11 domains paired with curated Skills and deterministic verifiers. Each task is evaluated under three conditions: no Skills, curated Skills, and self-generated Skills. We test 7 agent-model configurations over 7,308 trajectories.

Curated Skills raise average pass rate by 16.2 percentage points(pp), but effects vary widely by domain (+4.5pp for Software Engineering to +51.9pp for Healthcare) and 16 of 84 tasks show negative deltas. Self-generated Skills provide no benefit on average, showing that models cannot reliably author the procedural knowledge they benefit from consuming. Focused Skills with 2--3 modules outperform comprehensive documentation, and smaller models with Skills can match larger models without them.

Today we’re releasing OpenAI Privacy Filter, an open-weight model for detecting and redacting personally identifiable information (PII) in text.

It is designed for high-throughput privacy workflows, and is able to perform context-aware detection of PII in unstructured text. It can run locally, which means that PII can be masked or redacted without leaving your machine. It processes long inputs efficiently, making redaction decisions in a quick, single pass.

A good video is at https://www.theguardian.com/science/2026/apr/22/ai-powered-robot-beats-elite-table-tennis-players-milestone-robotics

Here we present Ace, to our knowledge the first real-world autonomous system competitive with elite human table tennis players. Ace addresses the challenges of physical real-time interaction through a new, high-speed perception system using event-based vision sensors4, and a new control system based on model-free reinforcement learning, as well as state-of-the-art high-speed robot hardware. Evaluated in matches against elite and professional players under official competition rules, Ace achieved several victories and demonstrated consistent returns of high-speed, high-spin shots. These results highlight the potential of physical AI agents to perform complex, real-time interactive tasks, suggesting broader applications in domains requiring fast, precise human–robot interaction.

Ace is equipped with a new perception system using event-based vision sensors as well as a control system based on policies learnt using deep RL. In contrast to earlier approaches using RL for robot table tennis, the control policies used by Ace are learnt using an asymmetric actor–critic architecture27,28,29, and the actions produced by the policies exist in an abstract space that is then mapped to a hard constraint for a convex optimization problem. This setup allows learning of collision-free, agile motions, addressing the full challenge of human-competitive robot table tennis.

The perception system uses a combination of conventional APS cameras for ball triangulation and EVS cameras for ball angular velocity estimation to infer the current ball state at high frequencies. The ball state is then provided to two different control components, depending on whether Ace is serving or in a rally. When serving, the robot performs a single-arm serve from a library of serve motions that were found using a genetic algorithm.

During the rally, a fixed deep RL policy (π′) is queried at 31.25 Hz using the robot joint states and the ball position and spin histories. The policy is sampled during the match from a bank of policies trained to perform different skills.

The actions (a) produced by the policy are mapped to a 32-ms segment trajectory, and a corresponding reset trajectory is calculated. If the robot has yet to hit the ball and no collisions are predicted, then the segment trajectory is executed by the robot interface; otherwise, a reset trajectory is executed.

The training of all policies is performed entirely in simulation with custom physics models, noise models and data-driven distributions of the initial ball state. Training is performed asynchronously with multiple instances of the training environment. To aid in the learning process, the critic () is provided with the true ball state, whereas the policy (πi) is given a history of noisy sensor measurements.

bicycle airtag holder

On Sunday, Lightning, a 5-foot-5-inch, bright red humanoid robot, completed a half-marathon in record time, almost completely on its own. But about 220 yards from the finish line, Lightning slammed into a barricade and fell over, almost as if it was giving up.  Despite the tumble, Lightning got back up with the help of its team and completed the race in just under 51 minutes.  Even with the disruption, Lightning outperformed the current world record holder, Jacob Kiplimo, who completed the half-marathon in 57 minutes, 20 seconds.

China has a large lead in robot manufacturing, shipping more than 1,000 humanoid robots, according to The Journal. No American company has shipped more than 500.

Why is there such a push for humanoid robots? Several countries are betting high on humanoid robots, but no country is betting more than China, as Straight Arrow News has previously reported. It makes sense since China has the most to gain from the technology.  China has maintained its position as the world’s largest manufacturer, but a massive issue could erode its dominance. Following decades of its One-Child Policy, China faces population decline. The sentiment among Chinese people has also exacerbated the problem, as fewer people are less eager to work low-paying manual labor jobs.  Robots could eliminate that issue for China, creating an entire labor force using its massive manufacturing infrastructure. China also understands humanoid robots would benefit every country and being the first country to perfect the tech would give the country massive leverage over others.

Many U.S. tech companies understand that advanced humanoid robotics will likely be a massive future tech market but they are focusing more on artificial intelligence. Many of them believe that building more advanced AI systems first allows them to create any technology afterward, only much faster.