Welcome to a new edition of misaligned bits, the (roughly weekly) newsletter from Misaligned where we sum up recent news and research, sometimes with a lighter touch.
As usual, we will mark all non-medium links with “➚” (external link) and all possibly paywalled links with “🔒”.
In our latest feature “The Great Escape Trick, Or: The New AI Opium For The Masses” we look at the language OpenAI and Anthropic have used to describe the recent security breaches, and why that choice of language matters.
Also, in a new edition of our “Papertrail” series, we look at how subtle LLM bias can shape narratives.
Artificial intelligence is now fuelling more than half of cybercrime in Africa, meaning more sophisticated attacks as financial losses mount, a new study has found. The study found that one of the fastest-growing threats is the creation of AI-generated synthetic identities that combine genuine personal information with fabricated ones. (➚🔒Bloomberg)
Turkey’s justice ministry is reportedly deploying AI to scrutinize legal documents for associations with groups designated as terrorists, a practice that experts fear will reinforce the judiciary’s documented pattern of human rights violations. (➚Stockholm Center For Freedom)
An article by Ian Lopez looks at the increasing use of chatbots in patient therapy and how some US states attempt to set limits. At least a dozen states have enacted legislation on AI chatbot protections for children, while Vermont and Rhode Island have instated AI therapy restrictions. Illinois directly restricts how licensed professionals utilize AI in healthcare, banning these tools from directly engaging with patients (➚🔒Bloomberg Law).
In the meantime, the Financial Times has reported that at least half a dozen senior NHS leaders have links to Palantir Technologies: The list shows Palantir recruited the people who ran NHS data before it won the £330M contract to hold NHS patient data (➚🔒Financial Times).
Anthropic has announced it will invisibly watermark all of Claude’s text and file output from 2 August to meet EU AI Act transparency rules: “Claude models launched on or after August 2, 2026 support marking at launch. We’re also working to add marking support to Claude models released before that date.” (➚Anthropic Support Statement)
The transparency obligation of the EU AI Act came into force on 2nd of August: The act requires that AI systems creating synthetic content, such as deepfakes, mark their outputs as artificially generated under Article 50.
Meanwhile, OpenAI has hosted an all-expenses-paid gathering of content creators at a New York state resort. This has resulted in backlash for some of the “brand influencers” involved. (➚Mashable)
SpaceX is planning to build a natural gas power plants for its future Texas chip factory that is supposedly to become the largest factory in the world.
Likely because Musk needs to destroy earth to get his moon and Mars colonization going, or more likely even because he does not give a shit about earth. Oh, btw, Musk recently (and privately) bought a gas turbine manufacturing company, so the self-dealing is all set up (➚🔒Bloomberg).
Meanwhile, Amazon is investing in a large-scale natural-gas-burning power plant as part of a huge data centre in Texas. The plant would use 35 natural-gas turbines to generate up to 7.65 gigawatts of power. It would be permitted to release 33 million tons of carbon dioxide a year, records show, more than any other power plant in the country (➚🔒New York Times).
Google’s proposed western Chesterfield data centre campus could disturb wetlands, streams and possible unmarked graves (➚Axios). Federal filings estimate the project could impact 4.8 acres of wetlands and over 2,800 feet of streams (➚Permit filings).
Chemical engineers at Stanford University in California have a used genome language models to design functioning genomes for bacteriophages. (➚The Guardian)
The scientists contend that this work raises “important biosafety, biocontainment and biosecurity considerations” and urged others doing similar work to “consult both safety and security professionals throughout the project”.
“Experimental tests yielded 16 functional genomes with diverse sequences, structures, and fitness profiles […] The generated phages were different from any known natural phages, exhibiting de novo mutations, divergent genes and regulatory elements, and variable genome lengths.” (➚Science)
A new study has found that AI models nearly erase female characters when they write kids stories about animals. (➚Eureka Alert)
Researchers gave six leading AI models variations on the same prompt they gave human participants: Across the responses, 57% of characters were either gender-neutral or ungendered, 41% were male, and just 2% were female
“Our findings point to a broader argument: neutrality bites. In other words, models that prioritize neutrality to address social bias may actually contribute to the erasure of marginalized perspectives and identities.”
(Note: Full study is behind paywall)
Imani Finkley, Yuanxi Li, and Melanie Walsh. 2026. Neutrality Bites: Gender Representation in AI-Generated Animal Stories. Proceedings of the 2026 ACM Conference on Fairness, Accountability, and Transparency. Association for Computing Machinery, New York, NY, USA, 2341–2358.
If you stumbled across this article on the web, subscribe to the “misaligned bits” newsletter and follow Misaligned on Medium, LinkedIn, Threads or Mastodon.
